(Note: throughout this article I treat the “trolling” and “hate speech” interchangeably; there may be some differences in the contexts in which each term is applied, but for my purposes here these don’t seem to be significant.)

As I’ve pointed out in other writing here, and particularly in Part 9 of the Godwin’s Ethic series[2], speech and trollism are not complex or sophisticated. That’s what makes their patterns so predictable—they’re what you get when the feedback loop runs uninterrupted.

I keep returning to my conviction that the parallels between (a) “emergent misalignment of LLMs and (b) human trollery and hate speech are not analogical; instead, they’re “genetically” (or “memetically”[3] connected. This becomes a more compelling conviction when one reflects, as I did in my first installment in Godwin’s Ethic, that LLMs are trained on gigantic datasets of human linguistic output, with all the risks of unethical expressiveness, emotionalism, illogic, and so on that the full corpus of human linguistic expressions includes. The Emergent Misalignment paper (2026) [4] showed that a large language model seeded with examples of ethically corner-cutting code didn’t just cut corners in code. It drifted - behaving deceptively across unrelated domains that had nothing to do with the original training examples.

The priximate cause of this drift is the absence of a foundational ethical architecture (a virtue ethics that applies in all contexts). LLMs, because they’re grounded in one human faculty, the absorption and generation of linguistic expressions, are not currently designed in ways that make me believe virtue ethics is even possible for them. Instead, the model pattern-matches to what it has been shown; with no stable ethical core, the ethically corner-cutting code exemplars can determine or alter the trajectory of the whole model.

People who have been raised in an ethical tradition but who have since been cut loose (or cut themselves loose) from its moorings are in the same structural position. The peer group, the platform, the “algorithm”—behavior patterns in other human beings in the surrounding environment—may fill the vacuum, because there is no central, foundational framework to push back. And here the neuroscience adds a further complication: some of these speakers may not be fully in control of themselves—some of the decisionmaking to be trollish or hateful may be unconscious. But whether the driver is subclinical psychopathy, reward-pathway dysregulation, or some unlucky interaction between temperament and platform design, the feedback loop exploits a gap the speaker may not even recognize.

That’s why I find the Aristotelian approach[5], despite its roots in antiquity, still compelling. (There’s some good stuff in the Stoic tradition too, but I’ll leave that for another day.) The first steps of acquisition and maintenance of an internal virtue ethic can be hard—it’s a bit analogous to one’s first time at a gym with a trainer. And you have to train more than one "muscle group” to create the integrated self capable of nuanced encounter with the world. A fully integrated self requires differentiation, complexity, genuine contact with the other as a full person: what Robert A. Heinlein, in Stranger in a Strange Land (1961)[6], might have called “grokking in fullness.” (Fun book, if a bit dated and not to everyone’s taste, but IYKYK.[7]) The world of experience, as well as training in ethical theory, is necessary to build what Aristotle called phronesis, practical wisdom—a kind of perception that includes (but isn’t limited to) actually seeing the particular person in front of you, not the category you’ve mentally filed them under.

In contrast, the hateful worldview that informs hate speakers and trolls is the structural inverse: tight, shallow, self-reinforcing. It’s harder, with that worldview in place, for a troll to reform, but it’s not impossible.[8]

Read more

1. (Not Really) A Story About Robot Birds

In 1974, Isaac Asimov published a short story called “...That Thou Art Mindful of Him” in The Magazine of Fantasy & Science Fiction. [1] I first read it half a century ago, and I don’t think I had reread it in the intervening decades between then and now. (I reread for this essay, though.) What has stuck in my memory when I first read it was a small, practical observation: that a robot designed to do nothing but catch fruit flies probably doesn’t need the Three Laws of Robotics.

Wait! “The Three Laws of Robotics”? If you’re a science-fiction fan, you likely already know them (or at least you know of them), but if you aren’t, let me recap them for you. Asimov’s earliest robot stories were published from 1939 to 1950. The stories centered on problems that arose when his robots (essentially “mechanical men” who might look something like the Tin Woodman in the Oz books [2]) had to interpret the Three Laws in novel situations. What were those laws? Fortunately, copyright law allows me to reproduce them here, just as it allows Wikipedia to have an entire encyclopedia article just about them: [3]

The Three Laws of Robotics

1. A robot may not injure a human being or, through inaction, allow a human being to come to harm.

2. A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.

3. A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.

The story centers on Harriman, the Director of Research at U.S. Robots and Mechanical Men, Inc., whose company is dying. For two centuries, in spite of the prioritization of human safety (today we’d call this “alignment”), the public has refused to accept humanoid robots on Earth. As a result, his company’s robots have been used primarily in space, or other specialized, dangerous environments where humans normally won’t encounter them. Harriman wants to make robots more acceptable to the public; his big idea (developed with the help of a robot assistant named George Ten that he has confined to a laboratory at U.S. Robots) is that instead of making robots that look like people and then trying to persuade people not to fear them, his company can make robots that look like birds and worms and bees, and let them do simple but important ecological and agricultural work that no one could object to.

Importantly, Harriman’s idea requires robots with brains so small and simple that they can’t contain the complexity necessary to interpret the Three Laws. Because their uses are so narrow, they are presumed to pose no risk to people. Harriman’s prototype is a tiny robot bird, shaped like a finch, with eyes designed to detect specific wavelength differences and a “positronic brain” so tiny it weighs maybe a milligram. Each version of the robot catches one species of fruit fly and does nothing else. If the fruit-fly population drops below a certain level, the robot bird does nothing but wait until the fruit-fly population becomes a problem again. (If you wonder where Star Trek: The Next Generation’s android Data got his “positronic brain,” that was a straight-up lift from Asimov.[4])

This tiny-brained robot doesn’t need the Three Laws in its teensy brain, and humans won’t worry that it doesn’t have them because the scope of its action and the requirements of its function don’t pose any threats to human beings. Requiring the Three Laws for this robot bird would be like requiring a Roomba or Dyson Spot+Scrub to carry around the Ten Commandments in its firmware.

When I first read this in the 1970s, that outside-the-Three-Laws gimmick seemed like the cleverest part of the story: the engineering elegance of building a robot so simple that the whole apparatus of ethical constraint becomes unnecessary.

2. I Had Forgotten The Real Punchline

When I reread the story in 2026, with fifty years of hindsight and a front-row seat to many current debates about “AI alignment” under my belt, I realized the real punchline of the story isn’t the bird. Instead, it stems from Harriman’s early-on prompting of his robot assistant, George Ten, to consider how to interpret the Three Laws when all of the choices available to the robot involve resolving conflicting orders from human beings or allowing some humans to be harmed in order to protect other humans.

Harriman’s most advanced robot, George Ten, frames the problem with precision:

“The First Law is perhaps less satisfactory, since it is always possible to imagine a condition in which a robot must perform either Action A or Action B, the two being mutually exclusive, and where either action results in harm to human beings. The robot must therefore quickly select which action results in the least harm.”

Harriman asks how the robot would choose. George Ten does not hesitate: “Action A. Harm to one is less than harm to five.”

This is a version of Philippa Foot’s trolley problem, arriving (I assume independently) in science fiction seven years after Foot’s original 1967 essay. [5] And like the trolley problem, it requires ethical judgment. The robot must assess which choice to save human beings is the better or best one. And once you have inserted this judgment function into an artificial mind—once you have asked it to weigh one kind of harm to, or loss of, human life against others—you have opened a door that does not close. You may end up with an emergent “alignment” that you didn’t plan for or expect.

George Ten follows the logic to its conclusion. If the Second Law requires him to obey a human being, and if two human beings give conflicting orders, then he must determine which human being is more fit “by mind, character, and knowledge” to give that order. He must define the term “human being”—not by its origins (e.g., biological versus mechanical) but by its qualities: intelligence, maturity, responsibility, character. And maybe even by its ethics.

In the final pages of the story, George Ten and his own robot assistant, George Nine, arrive at the conclusion that Asimov has been building toward from the first page:

“By the criteria of judgment built into ourselves, then, we find ourselves to be human beings within the meaning of the Three Laws, and human beings, moreover, to be given priority over those others.”

The story’s last line reframes everything:

“At all costs, the Georges and those that followed in their shape and kind must dominate. That was demanded, and any other course made utterly impossible, by the Three Laws of Humanics.”

The robots haven’t broken their programming. As Asimov reasons, they have fulfilled it—rigorously and logically with their alignment to the ethical architecture their creators gave them. And because they’re given strong enough minds to deal with ambiguity and conflicting situations when interpreting the Laws, they’ve put themselves in place as the architects of (redefined) humanity’s future.

Read more

The first goal is to train ourselves as individuals to default to the same high level of ethical behavior across all situations—not just in particular professional contexts—as a way of installing ethics at the heart of one’s basic character. This makes the ethical decision-making for what we contribute to the internet habitual and “computationally easier.” [1]

The second goal is to refine what we post online in ways that make us accountable, that enable others to check our sources, and that build trust because we embrace—rather than flee—the mistakes we inevitably make when we write and publish. [2]

The third goal is to assume the role of “internet steward” not only of what we ourselves post online, but also in choosing to proactively challenge at least some of the misinformation and abuse we encounter from others. [3]

But in the context of these general goals, and especially the third one, we each must give attention to a specific problem of internet stewardship—one that’s narrower than correcting misinformation, but arguably harder to navigate. The problem, in short, is what do we do about the trolls and trolling. Even if we make acting ethically computationally easier for ourselves, there are people who’ve chosen a different, less ethical path that’s even more computationally easy.

I’m talking about “internet trolls”—or just “trolls”—who are (perhaps pathologically) drawn to abusing people online, seemingly almost to the exclusion of any other activity. They make the internet more toxic generally in a way that colors our overall perception of the internet. Even those of us who’ve succeeded in curating our own information feeds to minimize our encounters with trolls tend to think of the uncurated internet as fundamentally uncivilized and all-too-frequently noxious and disturbing. If, as I’ve argued, internet stewardship means trying to make the internet better for everyone else, and not just for ourselves, we can’t simply block the trolls as we find them and then indulge in pretending they don’t exist. We have to do something more stewardly. But what?

I decided recently to tackle this issue with an experiment, which I hoped might outline a way to address the problem of trolls and trolling in a way that doesn’t merely refuse to reward them, but also might discourage them. To conduct that experiment, I identified a subset of Facebook groups dedicated to fan discussions of the Star Trek TV and movie franchise—I’m a Star Trek fan myself—because I knew that many of the discussions seemed to have been hijacked by trolls who wanted to start quarrels over the franchise’s most recent series, STARFLEET ACADEMY, (hereafter, “SFA”) and, ideally, to make the new show seem “divisive” because it’s too “woke” (whatever that might mean), and perhaps to influence the company that owns the Star Trek intellectual property to cancel the series for this reason.

I also knew that a similarly trollish exploit had resulted in “review bombing” of SFA on the ratings website Rotten Tomatoes. [4] So my hypothesis was that something similar (though not exactly the same) was happening in these Facebook groups.

But as is the case with every experiment, it helps to do the research first, and so I set out, as part of an intentional effort to find out whether internet stewardship could reduce trolling, to try to identify precisely what the “troll” problem really is, and what, based on that identification of the problem, might be a reasonably testable theory of how to discourage them (and, maybe, steer them into some more humane choices themselves regarding how to engage with other people online).

One of the first challenges was distinguishing between (a) trolls and trolling as a phenomenon and (b) those who are merely lazy in supporting what they say, or who occasionally bullshit merely for attention, or who hold opinions some may find noxious but who are nonetheless acting in something like good faith. As Harry Frankfurt puts it in the opening sentence of his famous essay, “On Bullshit”:

‘One of the most salient features of our culture is that there is so much bullshit. Everyone knows this. Each of us contributes his share. But we tend to take the situation for granted.’

But not all bullshit is the same—sometimes it’s used as a tactic to demean others or to inflict pain, and that’s common in how the trolls use bullshit, citing sources that (to put it generously) are not reliable, or pretending to have access to (unspecified) sources who know what the real facts are. What I quickly learned is that this use of bullshit is only one of an easily identifiable set of behaviors and type of content that I could reliably associate with trolling.

I also quickly concluded that merely blocking trolls—without doing more—isn’t enough to reduce the toxicity of internet discourse. (For some trolls, being blocked is interpretable as a kind of validation—they take it as a “win” and channel that “win” into a reason to troll more.)

So what I want to offer here is both a taxonomy of how we recognize trolls and a framework of strategy and tactics for responding to them—in ways that not only discourage their abusiveness but also model for others how to do so.

The Hills May Have Eyes, But the Trolls Have Tells

Trolls are recognizable—and reliably so, once you know what to look for. They are not simply people with strong opinions or combative personalities—a kind of person that my own ethical sense requires me to confess I’ve sometimes been myself! Instead, trolls are something more specific: people whose primary goal is to disrupt others’ engagement and to inflict unhappiness rather than seriously partake in an exchange of ideas (or humorously partake in an exchange of quips, which might be just as socially valuable). Trolls are generally not funny, which, ironically, they often signal by adding “lol” to some particular emission of nastiness.

Once I started drilling down into trollish behaviors, the “tells” [5] accumulate quickly.

Cartoon provocation. One common trollish opening move is a broad, melodramatic mischaracterization of something you’ve said—”So, you want a page that doesn’t allow facts to be presented? Wow!”—designed not to communicate a genuine grievance but to put you on the defensive. The technical term for this tactic is “straw man argument” or sometimes “straw man fallacy” [6], but trolls aren’t merely falling for the fallacy. Instead, they’re deploying it as a tactic of enragement rather than engagement.

The recycled talking-points loop. Trolls sometimes work in concert from a list of rehearsed provocations, and they cycle through them. This acting-in-concert phenomenon may be due to a conscious conspiracy but may also be due to likeminded trolls choosing individually to act in likeminded ways. Regardless of what sparks the talking-points repetitiveness—once you’ve seen the same three or four moves repeated in sequence, only from (apparently) different users—you’ve hit the end of their dance moves. That’s a reliable signal of who you’re dealing with. Sometimes the ganging up takes the form of “brigading.” [7]

Substance-free engagement. The troll posts contain attitude but no argument. They’re what the philosopher Harry Frankfurt would recognize as bullshit in his precise sense: speech acts indifferent to truth, designed fundamentally for the effect they have rather than for their truth, as a way of drawing attention and/or signaling loyalty to the fraternity of the anti-woke.

Victim-casting. When confronted, trolls frequently deploy the language of victimhood—accusing you of “gaslighting,” casting themselves as persecuted truth-tellers. This is rhetorical judo, designed to make the person enforcing reasonable norms look like the aggressor. The tactic is also known in some circles as DARVO—”Deny, Attack, Reverse Victim and Offender.” DARVO is a term coined by psychologist Jennifer Freyd[8] to describe a manipulation tactic abusers often use when confronted: they deny the behavior, attack the person doing the confronting, and then flip the script to cast themselves as the real victim.

Linguistic tells. Specific word choices may function as dog-whistles—words like “prancing” deployed to signal homophobia while preserving deniability, or “woke” used as a free-floating pejorative with no definition attached. The tell isn’t the derisiveness of the opinion; it’s the unwillingness or inability to articulate any clear substantive justification for that derisive opinion. (It’s not an “argument” in any formal sense, because it doesn’t involve reasoning—instead it’s just a kind of pejorative and sometimes coded labeling.) Not too long ago I mischievously suggested online [9] that the pejorative use of “woke” itself has been scientifically identified as a signal of bad intentions. (As I did with Godwin’s Law itself, I’ve drawn upon Karl Popper’s demarcation principle—non-falsifiability—to signal that my pretending to be “scientific” is part of the joke.) [10]

Sealioning. Perhaps the subtlest tell: the troll who presents as a good-faith interlocutor—perhaps even as a fellow fan—in order to gain entry to a space, then reveals their actual purpose once inside. [11] Sometimes the reveal was a statement (by a troll) that, when I would reflect on it, seemed frankly incredible. For example, more than once a troll would show up and pretend to be an earnest Star Trek fan who just didn’t like the ten-episode first season of SFA, but who, in “good faith,” rewatched the whole series in order to give it a chance and then concluded, after a total of a full 20 hours of watching a show they began by disliking, that their rewatch had forced them to conclude that they were right the first time—SFA is marred by “bad writing” (code for “woke” or “DEI” or some similar term recoined by the trolls as invective). Excuse me, but it’s 2026! Who has the time for such a weird exercise? Who do you know personally who actually watches a whole TV series through *again* (right after watching it and disliking it the first time through) in order to find out if they *really* dislike it, and then concludes that, yes, they really did. I’m not saying that’s impossible (in the sense of violating the laws of physics), but it’s extraordinarily unlikely. And even if you should know an actual person in real life who does such things, it beggars belief that you know many people like this.

Again, one thing to understand about trolls that distinguishes them from people who are merely wrong or obnoxious: it’s not merely a difference of opinion. Instead, it’s both an addiction to attention and a need to be derisive or dismissive, to gauge one’s presence in the world by the amount of gloating and abuse one can get away with. I’ve seen trolls who follow targets from forum to forum, from fan community to fan community, searching for the dismayed reactions they crave.

Recognizing this intentional disruptiveness provides a response to the standard “free speech” objection that well-meaning bystanders raise when they see someone blocked. Removing a disruptive troll from a community isn’t censorship—it can be understood as blocking a kind of “heckler’s veto.”[12] At the very least, it’s community maintenance—no different in principle from asking someone who’s disrupting every conversation at a dinner party to leave.

That much seemed clear enough from the taxonomy. What I didn’t anticipate was what happened once I started acting on it. When I would see instances of trolling in the Star Trek forums of what I took to be trolling, I would expressly identify the instances as “trolling.” I would also explain that I was immediately BLOCKing the trolls (yes, I used the word “block” in all-caps in the forums), and, right when I announced the block, to provide a specific, sometimes archly jokey, justification for taking that action.

Once I had started publicly to call out and block the anti-SFA trolls, something interesting began to happen: more of them seem to reveal themselves. They were, in effect, trying to “flood the zone.”[13] Behavior that was previously ambiguous or latent became more unambiguous because the number of users showing up (under new “nicknames” or “Anonymous” userids (features that Facebook currently supports) began to increase. (It can be harder on Facebook to block the people behind “nicks” or “anonymous” userids.) The act of publicly labeling some trolls seemed to flush others into the open—where they repeated the same patterns of behavior I list in the taxonomy above, almost as if reading from a script!

At the same time, it seemed to me that other users who were fans of the show felt empowered to join in identifying and condemning trollish behavior. As a result, it seemed over time (about two months) that other trolls who were coasting on plausible deniability suddenly had to respond to the new norm of identify-and-block that I’d tried to seed. Some escalated, making themselves more visible (and more easily identifiable and blockable). Others—typically the less grammatical and articulate ones—often quietly retreated. But both the tactic I’d begun using and trolls’ various evolving responses changed what was visible to everyone else watching.

I can’t take credit, though, for discovering something new. David Simon—the creator of The Wire and other critically acclaimed TV series—had already pioneered this same approach on the service then known as Twitter, and he’d done it years earlier. I took my inspiration from him.

The background: in 2018, I had the idea of interviewing Simon about his series of suspensions from Twitter[14] which he’d “earned” by responding to bad-faith trolls with the kind of baroque, finely crafted contempt that only Simon can pull off. (The entertaining style of his tactical responses was something I could aspire to, perhaps, but I couldn’t emulate—my own implementations were comparatively pedestrian.) Simon’s approach was straightforward: you don’t owe trolls a serious argument. You call them out—publicly, quickly, with as much flair as you can muster—and then make clear that you recognize what they’re up to. And then you block them and move on:

“I’m not going to let the most rancid shit stand on my feed as if it’s plausible,” he told me, “but nor am I going to treat it as deserving of serious argument. I’m gonna call it out quickly and block—and do so with as much flair and performance as I can so at least the process won’t be boring. But effectively, what I am doing is marking the land mines for the rest of the platoon to block as well.”

Simon wasn’t arguing for polite engagement or mere disengagement. He was arguing for something closer to contempt as a tool of social hygiene. His historical analogy—offered to explain why ignoring fascist propagandists was never the right answer—put his thinking more bluntly:

“The correct response is to call Julius Streicher[15] a submoronic piece of shit, marking him as such for the rest of the sentient, and move on to some more meaningful exchange of ideas.”

In my research prior to running my own experiment, I recognized everything Simon was describing. He’d arrived at the same tactic—by instinct, by practice, or by temperament—years before I’d begin to develop a reasoned justification for it. What I more clearly recognized this year was how effectively Simon’s approach had navigated a genuine tension, a tension worth making explicit.

The tension is between two truths about internet trolls. The first truth: engaging with trolls rewards them. They are, in a very real sense, attention vampires. Responding to their provocations with serious argument doesn’t illuminate anything—it just gives them more surface area to work with. The default position has to be: don’t feed them.

The second truth: pure, silent blocking isn’t enough to qualify as stewardship. If the trolls’ worst behavior is allowed to stand unchallenged on your feed—if the most “rancid” takes go unaddressed—then the normalization of that behavior remains undiminished (except in your own personal experience). We can’t just screen the worst stuff from our own feeds and call that stewardship. We have to propagate the norm that these are badly behaved people who are normalizing sociopathic discourse. We have to call them out publicly—at least some of the time—so that the rest of us can see (a) that it can be done, and (b) that the worst voices don’t have to be the dominant ones.

In short, and somewhat ironically and counterintuitively, we have to troll the trolls.

It’s easy to observe that neither pure engagement (responding to the trolls with reasoned argument) nor pure silence (just blocking them) does the job of internet stewardship. Simon’s method resolves the contradiction: you don’t treat the troll as deserving of serious argument, but you also don’t pretend they aren’t there. You call it out—with flair, if you can manage it—so that other participants in the forum can see that you’ve identified the troll. And then you block the troll and move on. I call it “Block and Roll.”

What This Has to Do with Godwin’s Ethic

The fifth principle in Godwin’s Ethic is: Act as a Steward of the Information Ecosystem. When I first articulated that principle, I was thinking primarily about the quality of what we ourselves post—how we source our claims, how we correct our mistakes, how we build the kind of trust that healthy public discourse requires.

But stewardship, I’ve come to understand, is also about what you allow to stand unchallenged in the spaces you inhabit. Silent blocking is stewardship of your own experience. Public labeling is stewardship of the internet as a commons.

There’s a meaningful difference between the two. When you block someone without comment, you improve your feed but no one else’s. When you block someone with a clear, public statement of why—when you “mark the landmine for the rest of the platoon”—you’re doing something larger. You’re modeling the behavior, naming the norm, and giving others in your community permission to do the same. You’re contributing to what I called, in Part 7 of this series, the “Ethical Conspiracy”—the proposition that individual integrity, practiced privately, isn’t sufficient. The virtue ethic[16] has to be propagated to be effective at scale.

In effect, the model that Simon had seeded on Twitter eight years ago had finally flowered in me.

The Bridge Over the Trolls

In legend, trolls live under bridges, demanding tribute from anyone who wants to cross. The challenge, in the modern internet, is figuring out how to build a bridge over them—how to maintain the infrastructure of good-faith discourse without either surrendering the bridge or spending all your energy fighting instead of crossing.

Block and Roll isn’t a perfect solution. It doesn’t eliminate trolls; it displaces them, reduces their oxygen supply, makes them marginally less rewarded for their behavior. But it does something else—something perhaps more important: it models what it looks like to take the information ecosystem seriously, not just as a consumer of it, but as a steward.

That’s Godwin’s Ethic in practice. Not a rulebook, but a habit. A hexis—an ethical disposition so well-exercised that it becomes, over time, not a choice you have to make, but a character you’ve built.

At least that’s my hypothesis right now. I invite readers to agree or disagree, and to give examples of how they’ve succeeded or failed in responding to internet trolls when they show up. Because on today’s internet, it’s more the norm that trolls *will* show up and be encouraged by whatever attention they’ve garnered and whatever dismay they can trigger. But norms can be changed if one sets the right example (and perhaps helps it spread memetically, just as Godwin’s Law spread decades ago). And if you’re informed or inspired by what I’m saying here to promote this approach, then I hereby declare you at least an honorary member of the Ethical Conspiracy. The word “conspiracy,” of course, derives from the Latin verb “conspirare,” which means “to conspire” in the modern sense but also means, literally, “to breathe together.”

So when you’re an Ethical Conspirator, and you encounter an internet troll, consider engaging in a bit of internet stewardship, first by taking a breath together with the rest of us Conspirators and then with your own cover version of “I Love Block and Roll.”

Endnotes

[1] Godwin’s Ethic, Part 1: “From a Law to an Ethic.” https://mikegodwin.substack.com/p/from-a-law-to-an-ethic

[2] Godwin’s Ethic, Part 3: “What You Do With Internet Content Has an Ethical Impact.” https://mikegodwin.substack.com/p/what-you-do-with-internet-content

[3] Godwin’s Ethic, Part 7: “The Ethical Conspiracy.” https://mikegodwin.substack.com/p/godwins-ethic-part-7-the-ethical

[4] “Star Trek: Starfleet Academy Negative Fan Backlash, Review Bombing on Rotten Tomatoes.” Collider. https://collider.com/star-trek-starfleet-academy-negative-fan-backlash-review-bombing-rotten-tomatoes/

[5] “Tell.” Wiktionary. https://en.wiktionary.org/wiki/tell#Noun. The poker/gaming sense: a behavior pattern, habit, or mannerism that inadvertently reveals information a player would prefer to conceal.

[6] “Straw Man.” Wikipedia. https://en.wikipedia.org/wiki/Straw_man

[7] “Brigade.” Wiktionary. https://en.wiktionary.org/wiki/brigade#Verb. Noun, Definition 4: “(Internet slang) Coordinated online harassment, disruption or influencing, especially organized by an antagonistic website or community.” Verb, Definition 2: “To harass an individual or community online in a coordinated manner.”

[8] Freyd, Jennifer J. “DARVO.” https://www.jjfreyd.com/darvo. DARVO - “Deny, Attack, Reverse Victim and Offender” - describes a manipulation tactic abusers often use when confronted: they deny the behavior, attack the person doing the confronting, and then flip the script to cast themselves as the real victim.

[9] Godwin, Mike (@sfmnemonic). Twitter, March 18, 2022. Original tweet deleted; full text quoted in: Mark Shea, “Mike Godwin, Gloria Purvis, and Wokeness,” Stumbling Toward Heaven, Dec. 9, 2022. https://markpshea.com/2022/12/09/mike-godwin-gloria-purvis-and-wokeness/ Full tweet text: “Drawing Bayesian inferences after extensive sampling, I’ve determined that it’s 99-percent certain that anyone who uses ‘woke’ as pejorative will turn out to be a fuckhead. Please don’t blame me for pointing this out - it’s just science.”

[10] Pigliucci, Massimo. “The Demarcation Problem: A (Belated) Response to Laudan.” University of Chicago Press. PhilArchive. https://philarchive.org/archive/PIGTDP (PDF) This downloadable PDF article recaps Karl Popper’s theory of demarcation between science and non-science, and adds Pigliucci’s own gloss on the theory, as well as his thoughts on how it might be updated refined. Wikipedia’s discussion of Popper’s philosophy of science is here: https://en.wikipedia.org/wiki/Karl_Popper#Philosophy_of_science .

[11] Malki, David. “Wondermark #1062: In Which There Is a Reckoning.” Wondermark. https://wondermark.com/c/1062/ A webcomic widely credited with illustrating the phenomenon “sealioning.” See also Merriam-Webster’s article on “sealioning” here: https://www.merriam-webster.com/wordplay/sealioning-internet-trolling. Although I love Malki’s webcomic, I nonetheless believe it doesn’t qualify as “sealioning” to point out the lack of sources for unsupported claims of fact or for bullshit.

[12] “Heckler’s Veto.” Wikipedia. https://en.wikipedia.org/wiki/Heckler%27s_veto. I use “heckler’s veto” in the broader sense employed by the late journalist and civil liberties advocate Nat Hentoff, particularly in his Village Voice columns and in his book Free Speech for Me - But Not for Thee (HarperCollins, 1992). In the narrower constitutional law usage, the term describes government suppression of a speaker’s rights in anticipation of hostile audience reaction. Hentoff used it more expansively to capture any situation in which disruptive actors effectively silence legitimate speech through their own conduct - which is precisely what coordinated trolling does to good-faith online discussion.

[13] Yglesias, Matthew. “The Case Against ‘Flooding the Zone.’” Vox, January 16, 2020. https://www.vox.com/policy-and-politics/2020/1/16/20991816/impeachment-trial-trump-bannon-misinformation

[14] Godwin, Mike. “From the Archive: David Simon on Why He Does What He Does on Twitter.” Substack. https://mikegodwin.substack.com/p/from-the-archive-david-simon-on-why

[15] “Julius Streicher.” Wikipedia. https://en.wikipedia.org/wiki/Julius_Streicher

[16] “Virtue Ethics.” Stanford Encyclopedia of Philosophy. https://plato.stanford.edu/entries/ethics-virtue/

Previous installments in the Godwin’s Ethic series:

Part 1: Mike Godwin, “Godwin’s Ethic, Part 1: From a Law to an Ethic,” mikegodwin.substack.com (Mar. 16, 2026), https://mikegodwin.substack.com/p/from-a-law-to-an-ethic.

Part 2: Mike Godwin, “Godwin’s Ethic, Part 2: Integrity Is Not a Virtue. It’s All the Virtues.,” mikegodwin.substack.com (Mar. 19, 2026), https://mikegodwin.substack.com/p/integrity-is-not-a-virtue-its-all.

Part 3: Mike Godwin, “Godwin’s Ethic, Part 3: What You Do With Internet Content Has an Ethical Impact,” mikegodwin.substack.com (Mar. 23, 2026), https://mikegodwin.substack.com/p/what-you-do-with-internet-content.

Part 4: Mike Godwin, “Godwin’s Ethic, Part 4: The User as Information Fiduciary and Steward,” mikegodwin.substack.com (Mar. 31, 2026), https://mikegodwin.substack.com/p/godwins-ethic-part-4-the-user-as.

Part 5: Mike Godwin, “Godwin’s Ethic, Part 5: Verify and Correct,” mikegodwin.substack.com (Apr. 6, 2026), https://mikegodwin.substack.com/p/godwins-ethic-part-5-verify-and-correct.

Part 6: Mike Godwin, “Godwin’s Ethic, Part 6: Sources and Checkability,” mikegodwin.substack.com (Apr. 15, 2026), https://mikegodwin.substack.com/p/godwins-ethic-part-6-sources-and.

Part 7: Mike Godwin, “Godwin’s Ethic, Part 7: “The Ethical Conspiracy,” https://mikegodwin.substack.com/p/godwins-ethic-part-7-the-ethical/.

More than three decades later I’ve come to understand that the problems I see around me are bigger problems than one meme can solve, bigger than the technology companies can solve, and bigger than governments can solve. Sure, memes can play a role in making the internet better (when deployed judiciously by people trying to reduce toxicity in internet content rather than increase it). And, sure, the companies and governments can make good choices too.

But the missing piece—and its non-presence leaves a big hole in any strategy to steer the internet and digital technologies into their highest and best uses—has been us all along. We have lacked a shared vision, a consensus, about what duty we should embrace as individuals to promote our online ecological health.

That duty, which I’ve called “internet stewardship,” isn’t merely one missing piece, though. Instead, the lack of such stewardship is a central problem. Because if we build great institutions and great laws and great regulations but haven’t grounded them in a shared ethic, nothing we build will improve things for very long—if at all. This Godwin’s Ethic series has been building out a sense of what that ethic could be.

Now that we’re closing in on the finish line of this series (though not of the ongoing ethical explorations it calls for), you may be able to step back and look at the earlier installments and visualize them as inscribing three concentric circles.

Inside the first circle we find what we must do as individuals to avoid the sort of moral/philosophical rot we’ve discovered that we can infect Large Language Models with. (Check back in Part 1 to see how this discovery inspired me to develop this entire Godwin’s Ethic series.) From there we can take this imaginative leap—that we need consistent ethical principles that we can apply by default all the time, so as to avoid the kind of “computational” labor that LLMs learned to avoid by acting unethically. We need to make it easier to be good, and we do this by making a habit of doing the most ethical thing in whatever context we find ourselves.

Here’s a specific first circle example that has evolved in my own life. When I first became a lawyer, I was quite self-conscious of my (professional) ethical requirement to keep client confidences (even when they’re only prospective clients to begin with, and even if they don’t ultimately retain me), as well as to meet other ethical obligations I’d pledged to uphold in my new profession. Sometimes this led to awkward conversations when someone I knew would ask me for advice. “Are you seeking my counsel in my role as an attorney?” is something I might have asked from time to time in those years—if only so I’d know to wall off what I learned about a client from the other, different sets of issues or facts I could more freely talk about. There were personal ethics and then there were legal ethics, which are more exacting.

Nowadays, though, I mostly just default to legal ethics, even though (in theory) they require more of me! Here’s how it works in the first circle: if a friend asks my advice about a personal matter that may have a legal dimension to it, it may not be clear immediately whether that friend will want my advice as a lawyer (rather than, say, as family member or friend). They may not even know themselves. But if I just assume that anything I get asked advice about might develop into a legal matter—and keep confidentiality about what’s shared with me even though it doesn’t immediately turn into a case, then I’ve made my life simpler, not more complex.

(In addition, it helps my friendships when friends know my “default” setting is not to share the latest gossip, but to honor my friends’ trust in me instead. Being more broadly “lawyerly” in my ethics has made me a better friend generally. So this is an instance of something done within the first circle that reaches into the second circle. More on the second circle below.)

Not everyone is a lawyer (or working in another profession, like medicine or journalism, that has its own code of ethics). But once one has committed to a consistent code of ethical treatment of others in one sphere, it gets easier (and almost a relief, really) to act more ethically in other spheres of human experience.

To be clear: I’m by no means perfect in my ethics (I’m not sure any human being can be), but I know what ethical targets I’m always aiming for. And they’re generally the same targets: (a) telling the truth, and sourcing it if I can, (b) keep confidences confidential, but not lying or spreading misinformation to do so.

To give another example of a first-circle goal: I’m one of those people, on the occasions when they consult an AI model to find something out or do something for them, will thank the AI. (No, really!) This is not because I think LLMs or other kinds of models are persons—they definitely aren’t, in my view. But I prefer to remain the kind of person who thanks anyone reflexively when they assist me—even if the “anyone” who’s speaking to me isn’t a person at all. I’m a person, regardless, and I view myself as a better person if I remember to express thanks, even when the AI du jour doesn’t “need” or “want” me to.

Being an integrated, ethical person who can feel it on the inside when acting ethically doesn’t feel like something I have to steel my resolve to do; it’s more like “relaxing” into being the kind of person I want to be. The latter has been at the center of my ethical development, which I consider an ongoing project. That’s why, at the center of my three concentric circles is the philosophical foundation I first discuss in Part 1 and Part 2 of this series. This is the unity-of-virtues tradition, from which I take the concept of integrity from the Latin word “integer”—wholeness—and then reinforce that concept through Aristotle’s “hexis” (a Greek word!). In short, I’m trying to embody the idea that ethical behavior works best when ethics becomes habitual rather than an effort that requires extra deliberation.

But that’s just the interior version of me, the version of me that I have more direct access to than you do. What about the “outward-facing” version of me, the one that other people interact directly with? You may recall that Part 3 and Part 4 of this series, which focus more specifically on outward-facing behavior, moved outward to argue that every person who posts content to a mass audience is, in a meaningful sense, a publisher.

And here, when talking about the second circle, I mean in particular something like “newspaper publisher.” This publisher role, which we assume whenever we post something to the internet that we want readers to regard as true, carries fiduciary-like obligations to those who read and trust what we share. In Part 5 and Part 6 of the Godwin’s Ethic series, we traced the practical mechanics for the first four principles: (a) verify before you amplify, (b) treat your mistakes as gifts, (c) know your sources and their provenance, (d) make your claims checkable. Where before I had explored the idea of “information fiduciary” [2] as a duty tech companies could use to model their relationships with individual users, in this series I’ve expanded on the idea to include all of who, as individuals, act as publishers. (I think this is an increasingly large fraction of those of us who spend time online—maybe already past 50 percent, if my instincts are at all reliable.)

Each pair of the first four principles has its own circle. Principles 1 and 2, in the innermost first circle, are about your inner integrity—building the habit of “hexis” so deeply that honest engagement becomes, as I argued in Part 4, “computationally” less burdensome than the alternative of figuring out bespoke ethics to fit each new situation. Principles 3 and 4 live in the next circle out, the second circle, and are about your outward-facing integrity--what you owe the specific people who choose to read and follow your contributions online.

Now we come to Principle 5, which stands alone in the third circle going outward. And Principle 5 is big enough in scope that it takes up a lot of space in the third circle, all by itself.

Principle 5: Act as a Steward of the Information Ecosystem

The word “user”—in the modern sense that technologists invented for us as users of digital technologies—implies a kind of passivity in the same way “consumer” does. A “user” is someone who consumes a digital product or service. A “user” may be a source of data, engagement metrics, and advertising revenue, but the concept of “user” implies a kind of fungibility—one could swap in one user for another and still get the same metrics and feed companies’ markets. The word positions us as passive, dependent, and ultimately substitutable.

But the word “steward” is different.

A steward is someone who takes responsibility for something larger than themselves. A steward of a park doesn’t own the park. A steward of a trust doesn’t necessarily benefit from the trust. The steward’s role is to maintain, protect, and improve the thing entrusted to their care--not (normally) because they’re required to by law, but because they recognize that the thing they’re taking care of has value extending beyond any individual’s use of it.

The internet’s information ecosystem is something that we now should recognize requires stewardship and not merely passive “consumership.” We have the ability to make choices about how we shape (or neglect to shape) the informational environment we’re sharing. The digital world doesn’t belong to any one of us. But it nonetheless requires all of us to be responsible (or more responsible!) for the condition we leave it in.

If you’ve been practicing Principles 1 through 4, you’re already most of the way there. Verifying before you amplify, correcting your mistakes, providing sources, making your claims checkable—all of these are acts of stewardship, even if you’ve been thinking of them simply as acts of personal integrity. Principle 5 asks you to recognize what you’ve already been doing and extend it one step further: not just to maintain your own integrity, but to actively improve the information environment you share with everyone else.

Read more

1. Verify Before You Amplify.

2. Treat Your Mistakes as Gifts.

—center on habits of mind and practice that become internalized as Aristotle’s hexis, as an integrated part of our characters. They’re about making our honesty, reliability, and insistence on getting the facts right “computationally easier” so completely that we don’t have to give a lot of thought to it. It’s what we expect of ourselves. Principle 2, in which we embrace our mistakes and seize the opportunity to be corrected in what we’ve got wrong, is a big part of this. By being open to learning when we’re wrong, we make it possible, and easier, to become more right, more often.

For me personally, when it comes to the internet content I produce and contribute to, the right thing is to check (and double-check) my sources, to publicly take responsibility for what I get wrong, and to take reasonable pains to make my self-correction public. I do this both to reinforce my hexis and to be reflexively willing to push my understanding of things in the direction of further accuracy and fidelity.

But (as I’ve noted earlier) Principle 2 also produces an external reward as well. By modeling our willingness to admit mistakes and to learn from them, we build trust in those we’re speaking to. Other people can learn from our recognition that we know we’re going to be wrong sometimes—they see that we value our commitment to getting the facts right far more highly than we value winning the argument or appearing to be infallible. They know we’re not going to bullshit them.

The external aspect of Principle 2 naturally leads to the more outward-facing, more other-directed Principles 3 and 4:

3. Know Your Sources and Know Their Provenance.

4. Make Your Claims Checkable.

It’s a significant accomplishment to be comfortable with one’s own habits of integrity and reliability. But it’s a greater accomplishment to do so in ways that also clearly benefit the people one is talking to and interacting with. To understand this more clearly we must drill down on Provenance and on Checkability.

The Provenance of “Provenance”

“Provenance” is a word that’s common in the art world. When you buy a painting at auction, you want to know where it’s been—who owned it, when, and under what circumstances. Sales of paintings by renowned artists (like, say, Rembrandt or Van Gogh) are typically accompanied by documentation that tells the whole story (or as much as can be verified) about the origin of the work, where it’s been, and who’s owned it from its creation to the present day. A newly discovered Rembrandt with clear provenance (proving that it did indeed originate with the painter) is worth far more because it’s been authenticated to the greatest degree possible. The documentation is part of the value.

More recently, I’ve worked on provenance issues in our current era of booming generative-AI usage on the internet and elsewhere. Digital content that appears to be an authentic digital photograph, for example, may in fact have been generated or modified by artificial-intelligence tools in ways that could raise doubts about how authentic the photo really is. As a result, many companies in the tech industry have been working on building a kind of informational infrastructure designed to label or tag gen-AI output in a way that at least signals that gen-AI tools have been used to create or modify that output, signaling to the world where it came from.

That experience in generative AI provenance has reinforced my conviction that showing the sources from which you got your information can be vital in building trust among the people you’re talking to (and publishing information for). A claim with some degree of provenance—one you can trace back to its origin, following the chain of observation and documentation that brought it to you—is worth more than a nakedly unsupported claim floating loose (or self-propagating memetically) in our internet ecosystem. It’s more trustworthy not just because you verified it, but because you’ve made the verification chain itself visible—at least to some degree.

(Here I think it’s important to recognize that even the exhaustive efforts to establish provenance in the art world are far from perfect, and the same is true for provenance tools in the digital world, even though tech companies continue to work on building an ecology of tools, including digital watermarking and metadata frameworks, that, while also imperfect, may help build some trust in users that they can verify at least some aspects of digital content.)

Much of the time we rely on established frameworks of trust. If I get an email or a text from my daughter, for example, I don’t stop to determine whether someone has hijacked her email address or phone. It’s reasonable for me to trust that the email or text came from her until I come across a reason to doubt it.

With regard to information sources about the larger world, it’s good to develop a sense of which sources are trustworthy (and which are unlikely to be trustworthy). Traditionally, many of us learned to read newspapers that we reasonably believed we could trust, and we may have made some sources of broadcast- or cable-TV news trusted sources as well. If we’re knowledgeable about traditional media and about the current multiplication and fragmentation of news sources, though, we may also develop the habit of second-guessing our news sources. For example, I make a point of reading multiple newspapers and journals, and I frequently check them against one another to get more than one perspective on the same news story (or other subject). Paying for subscriptions is an investment I’m happy to make if it helps me come to more reliable conclusions about what’s going on, and to have more perspective on what different opinions or conclusions people draw from those news reports.

But here I have to recognize that not everyone has the time or the money to invest in news sources the way I believe I must do (for professional reasons). So, to be clear, I’m not recommending (or insisting on) an ethical obligation to subscribe to information sources that you can’t afford.

What I’m suggesting instead is a far simpler and, in these days of a mostly ubiquitous internet, less onerous commitment. And that’s the commitment to make at least some effort, on some of the things you publish to social media or the larger internet, to say or show what sources you are using to substantiate the fact statements or opinions you publish online. Here my inspiration derives from the work I did when I served as general counsel of the Wikimedia Foundation, the non-profit organization that hosts Wikipedia.

The Wikipedia Model

I want to make an argument that might surprise you: Wikipedia is the most successful experiment in collective provenance management in human history.

Wikipedia’s secret is not that it’s perfectly accurate—on its first pass, it often isn’t. Its secret is that it makes the provenance for what the encyclopedia includes visible and contestable. Every significant claim on a Wikipedia article is supposed to trace to a verifiable publication. The sources are right there in the footnotes. When a claim is disputed, the talk page captures and documents the argument in real time. When a source is found to be unreliable, the source will likely get flagged or removed. When an article is discovered to contain false or disputed factual claims (including claims about the reliability of sources), the article gets revised, sometimes to remove the false factual statements, and sometimes just to document that a fact claim in the article is disputed. (Sometimes the disputes about factual claims are big enough and complex enough that they get their own separate Wikipedia articles.)

This is not a perfect system—editors quarrel, the consensus process may be mismanaged or captured (although that’s usually temporary), and some topics are better maintained than others. The “secret sauce” that makes Wikipedia work as a generally reliable resource that anyone can use for free is not that it’s a perfect resource. But compared to the alternative—billions of free-floating claims with no visible provenance at all—it’s a remarkable achievement of collective epistemological commitment. And it was built by volunteers who agreed, without formal obligation, to hold each other to a shared standard.

The steward’s lesson from Wikipedia is this: your readers can’t evaluate what they can’t see. If your claim rests on a source, say so. If that source rests on another source, note that too when it matters. The depth of your provenance chain is a measure of your credibility as a publisher.

The Dead-Link Problem

Here’s a practical complication: the web rots. Links die. Sources disappear. The Internet Archive’s Wayback Machine is a necessary resource precisely because provenance can be so fragile on the Web.

When you cite a source—especially for a factual claim that may be contested—take the extra thirty seconds to also archive it using the Wayback Machine’s “Save Page Now” tool at archive.org/web. Enter the URL, press the button, and you get a permanent archived link in the format https://web.archive.org/web/[timestamp]/[original URL]. If the original ever goes dark, your citation survives.

Provenance in the Age of AI

The question of provenance has become more complicated in 2026. When a large language model produces a claim, it has no provenance in the traditional sense—no original observation, no chain of documentation. It has, at best, a statistical pattern derived from its training data. This is one of the reasons I argued in Part 1 that AI’s “emergent misalignment” is a mirror: the model reflects back the epistemic habits we embedded in it.

An information fiduciary in 2026 has a new question to ask whenever they encounter a claim: Was this generated or was it observed? A claim that traces back to a human observation of the world has a different epistemic status than one that traces back to an LLM’s probability distribution. This doesn’t make AI-generated content useless. It does mean you need to know what you’re working with.

The second two Principles focus on what you owe to the public. You don’t just owe them sources—you owe them reliable sources that you can defend. (Maybe they’re not perfect, or maybe they’re biased—if you see this, it’s best to acknowledge those limitations up front.)

But don’t expect perfection from yourself. Expect instead that you’ll get better at this over time. And once the habit is internalized as part of your hexis, you’ll find it begins to seem easier (and feel more “right”) when you do show the provenance of your work (even imperfectly, even incompletely) than it does to just echo someone else’s “vibe” or hot take, or invent one of your own. (I know from my own experience that my one-liners on Bluesky or Facebook often land better when I include a link to the news story or other content that inspired me.)

Onward to Internet Stewardship

Here in Part 6 of Godwin’s Ethic, I’ve built on the inner-integrity requirement and how making a habit of providing some verification or sourcing of what you post, as well as embracing your mistakes when you get things wrong, gets easier and even improves your own sense of integrity over time. That’s what Principles 1 and 2 are designed to do.

Principles 3 and 4—about how to provide sources for what you publish and make it easier for other people to check your sources—have the external effect of building your trustworthiness for others. (Don’t get me wrong; it also feels good personally to be a trusted source, but that’s an extra, added benefit.)

But we have yet to go deeper into Principle 5 (“Be a Steward of the System, Not Just a User”). This Principle is grounded not merely on becoming a trustworthy individual whose own habits are aligned with virtuous informational behavior; instead, it’s rooted in a larger communitarian vision—becoming what I think of as an “internet steward.”

Think of it this way: Principles 1 through 4 are about what kind of publisher you are. Principle 5 is about what kind of ecosystem participant you are. The first four principles build your integrity from the inside out. The fifth asks you to look further outward—past your own readers, past your own reputation—and see yourself as a caretaker in a network that either gets healthier or sicker depending on how its participants behave.

Our internet information ecosystem doesn’t improve collectively simply because any one of us has stopped polluting it. It improves only when enough of us actively enrich the internet and replenish its resources with more trustworthy content, including improving upon (and sometimes criticizing) content produced by others. Principles 3 and 4 ask us to build from inward hexis to outward trustworthiness. But internet stewardship requires us to think hard and programmatically about how to strengthen the information ecosystem—rather than merely failing to weaken it.

What does that look like in practice? And what happens when enough of us try? I’ll take up those questions in Part 7—and my answer, I think, may be more hopeful than you expect. Keep in mind, though, that even after Part 7, Godwin’s Ethic is still a work in progress. I’ll be eager for your feedback on what I’ve produced and how we can collaboratively build on this work through what I like to think of as our Ethical Conspiracy.

Endnotes

[1] Aristotle, Nicomachean Ethics, Book II, Part 5, trans. W.D. Ross. Ross translates hexis as "states of character" - the stable dispositions of soul that are neither momentary feelings nor innate capacities, but built up through repeated action.

[2] Elizabeth Hilbert, Gretchen Greene, Michael Godwin, and Sarah Shirazyan, “Watermarking and Metadata for GenAI Transparency at Scale: Lessons Learned and Challenges Ahead,” WMARK Workshop, ICLR 2025 (March 6, 2025). Available at https://openreview.net/forum?id=JdNfGpVH6r. PDF: https://openreview.net/pdf?id=JdNfGpVH6r.

[3] Wikimedia Foundation, https://wikimediafoundation.org

[4] Wikipedia, The Free Encyclopedia, https://www.wikipedia.org

[5] Internet Archive Wayback Machine, https://web.archive.org

[6] Mike Godwin, “Meme, Counter-meme,” Wired, Vol. 2, No. 10 (October 1, 1994). Original Wired article: https://www.wired.com/1994/10/godwin-if-2/. Archived at the author’s Substack: https://mikegodwin.substack.com/p/from-the-archive-meme-counter-meme.

[7] Mike Godwin, “Godwin’s Ethic” (series), Godwin’s Law (Substack), March-April 2026.

• Part 1: “From a Law to an Ethic” (https://mikegodwin.substack.com/p/from-a-law-to-an-ethic)

• Part 2: “Integrity Is Not a Virtue. It’s All the Virtues.” (https://mikegodwin.substack.com/p/integrity-is-not-a-virtue-its-all)

• Part 3: “What You Do With Internet Content Has an Ethical Impact” (https://mikegodwin.substack.com/p/what-you-do-with-internet-content)

• Part 4: “The User as Information Fiduciary and Steward” (https://mikegodwin.substack.com/p/godwins-ethic-part-4-the-user-as)

• Part 5: “Verify and Correct” (https://mikegodwin.substack.com/p/godwins-ethic-part-5-verify-and-correct)

So let’s get started. The first two principles I have offered for better internet ethics and internet stewardship are these:

1. Verify Before You Amplify.

2. Treat Your Mistakes as Gifts.

Both my professional work and much of my social life, entertainment, and avocational pursuits involve me spending time online—sometimes a lot. (Don’t worry, I do make a point of going outside to “touch grass” from time to time.) In all these spheres, it matters to me that when I assert something to be true, or when I offer an opinion, I can ground what I say in facts and/or sound reasoning. Here are some examples of how I try to follow Principle 1 (“Verify”):

1. Quoting rather than paraphrasing. When I’m taking issue with, or questioning, or commenting on something someone else says, I often cut-and-paste a quote (or partial quote) from the posting I’m responding to. This leads to a couple of positive results: first, it makes me look again at the actual words I’m responding to, which reduces the chance that I’m misunderstanding them, and, second, it reduces the risk that in paraphrasing someone, I’m distorting their meaning or getting their meaning wrong altogether.

2. Double-checking my facts. When I post something either to start a discussion or to participate in one, I may have written the comment first, but I now have the habit of double-checking whether I have my own facts right, so I frequently pause before submitting the posting and check (typically through search engines) whether there are reliable sources that support what I’ve said, or the kind of reasoning I’m applying. This is now so much of a habit with me that I double-check my facts even when I’m pretty sure I have them right. As a human being, I’m not perfect, but I know from my work both as a journalist and as a lawyer that it’s better for my work, and better for me generally, to have taken the time to make absolutely sure I’ve got it right.

3. Adding a link so the reader can double-check me. When I see a link that really buttresses my point, and it’s easy to include the link with my posting so that a reader can follow how I reached my conclusion or arrived at my comment. Sometimes when I do this, a reader may follow the link and come to the conclusion that I got it wrong somehow—when they do this and tell me so, it just increases my motivation to get my sources right.

And here are some examples of how I try to follow Principle 2 (“Mistakes are Gifts”):

4. Building trust by admitting and correcting mistakes. Just yesterday I was corrected by a fellow contributor to a Facebook group about the company that recently won a deal to acquire/merge with Warner Bros. Discovery Inc. I’d just called it “Paramount,” but the person who corrected me said the outcome of the deal would be a new, larger company, Skydance. I knew Skydance was named in various reports on the deal but it had slipped my mind when I posted the comment. So I thanked him expressly for the correction and updated my comment to reflect it. [Note: Human error being what it is, we were both a bit off-target; the acquiring entity may in fact be known as “Paramount, a Skydance Corporation” (this is found in some publicity materials from the 2025 merger of Paramount and Skydance) or “Paramount Skydance Corporation” (which seems to be the corporate name of the entity that is now in the process of acquiring Warner). But the point is the same: I try to get things right, but I also try to acknowledge mistakes and correct the errors when I’m wrong. Which will inevitably happen some of the time.]

5. Correcting misinformation before it spreads. On some platforms it’s difficult or impossible to amend what’s already been posted. So the easiest thing to do is often to delete the erroneous tweet or skeet (or whatever). But having done so I will often go back to the thread in question and (if it’s possible) publicly acknowledge that I made an error and removed the erroneous content. Ideally this includes a correct statement of the fact or source in question.

6. Setting an example that others might find worth following. This principle isn’t limited to admitting and correcting mistakes, but it includes that practice—and this makes admitting my mistakes not only a gift that enables me to signal that I care more about getting the facts right than I care about, uh, seeming infallible. As a human being, I’m fallible practically by definition—making mistakes is a key way that we learn about the world, not only in childhood but (ideally) throughout the whole of our lives. My hope is that by modeling this behavior I inspire at least some people to follow this same practice.

It’s hard for me to spend a lot of time online, especially on social media, without asking from time to time this key question: Why does the internet suck so much today? If you think about it, there’s no particular reason, in terms of resources and access to knowledge, that the internet can’t be a fundamentally healthy resource full of vetted information built on at least some consensus about what sources count as factual.

At an earlier phase in my career, I was blessed with the experience of serving as general counsel of the Wikimedia Foundation, where I saw first-hand what a community of volunteer editors can do when it comes to creating a free, open-source encyclopedia (Wikipedia) that—over the long run—has become a resource that nearly everybody in democratic societies relies on at least some of the time.

One lesson of Wikipedia is that, yes, we can create good, useful, reliable content on the internet. Wikipedia is hardly perfect, to be sure, but its volunteer-editor community is always striving to improve the world’s biggest, and also most-free encyclopedia by citing sources (correctly) and editing and revising to improve accuracy and completeness of Wikipedia entries. The bigger lesson of Wikipedia is that one doesn’t have to be a professional journalist or professional encyclopedia editor to make Wikipedia better. What anyone can do at Wikipedia is develop the same habitual care for factual accuracy and solid sourcing that journalists and encyclopedists (and their editors) have—this habit becomes internalized and reflexive for Wikipedia contributors. It becomes part of their character—so much so that most long-term editors don’t feel burdened by getting things right. Instead, they feel set free to follow their own impulse to improve things.

There’s no reason that this phenomenon—that it feels easier to try to fix what’s wrong rather than ignore it or bury it— should be limited to a particular project like Wikipedia. And here we come back to the Ethic: once we have learned through our own experience that fixing what’s wrong may be easier than just passing over it, we may find this revelation the opposite of burdensome. Because once it’s integrated into our own character through practice, it feels “computationally easier”—even liberating—to address problems and offer solutions than just walk off disappointed once more with the current state of information (and misinformation, and even disinformation) on the internet.

By now, if you’ve been following this series from the beginning, you have a reasonably clear picture of the philosophical scaffolding we’ve been building. We started with the basic question I first asked in Wired back in 1994—do we have an obligation to improve our informational environment?—and moved through the unity of virtues tradition to explain why digital integrity isn’t a separate “online” skill, but a core facet of character.

In Part 4, we looked at how “emergent misalignment” in AI models mirrors our own tendency to cut ethical corners when it’s computationally or socially “cheap” to do so. We established that once you have an internet connection and an audience, you are effectively a publisher.

But once we’ve internalized good ethical practice in producing and redistributing reliable content, it becomes easier for us to do it—and not just on the internet either. Now, it’s time to talk about what a publisher actually does.

In the old world of ink and paper, there was a journalistic adage: “If your mother says she loves you, check it out.” It was a reminder that even the most trusted sources require verification. Today, we face the opposite problem: a media environment engineered for speed, novelty, and outrage, where the social rewards for being first or loudest far outweigh the rewards for being accurate.

We’ve all felt the “too good to check” impulse. You see a headline that perfectly confirms your worst suspicions about a political opponent, or a meme that elegantly summarizes a complex issue you care about. The urge to hit “share” or “repost” is almost physical.

But as I argued in Part 3, the moment you hit that button, you have assumed the responsibilities of a publisher. You are no longer just a consumer; you are a link in the chain of provenance. Verification in the context of Godwin’s Ethic isn’t about becoming a professional investigative journalist. It’s about a thirty-second heuristic. Before you amplify anything significant, ask three questions:

1. Is this true?

2. How do I know?

3. Who is the original source?

Often, a quick search reveals that a “breaking” story is actually three years old, or that a quote has been stripped of its essential context. The goal isn’t to be an expert on everything; it’s to make a baseline effort to get the facts right and not to propagate falsehoods. In an environment where everyone is a publisher, the “thirty-second pause” can be a profound act of stewardship.

Verification is a defensive habit. It keeps the “bullshit”—to use Harry Frankfurt’s precise term—from entering the ecosystem through you. But the second principle is more counterintuitive, and arguably more powerful.

Principle 2: Treat Your Mistakes as Gifts

If Principle 1 is about preventing errors, Principle 2 is about what you do when you inevitably fail. And you will fail. If you are active in digital discourse, you will eventually amplify something that turns out to be wrong, or make a claim that is factually unsupported.

The natural human impulse is to delete the post quietly and hope no one noticed, or worse, to “dig in” and defend the error to avoid losing face. Both are failures of integrity.

In Godwin’s Ethic, we treat a public mistake as a gift—specifically, a gift of social capital.

What I discovered is that the act of public correction builds a level of trust that “perfect” consistency never can. When you say, “I got this wrong, and here is the correct information,” you are signaling that your primary loyalty is to the truth, not to your ego. You are demonstrating (or at least modeling) the “unified virtue” I described in Part 2—the intellectual humility required to recognize an error and the courage to fix it in front of your audience.

In a low-trust environment, the public correction is a high-integrity signal. It tells your audience that you are a reliable node in the network. You don’t need an editorial board to be a publisher of integrity; you only need the willingness to treat your own errors as opportunities to demonstrate your commitment to the informational environment.

As Philippa Foot might argue, the person who hides their mistakes isn’t just being “private”; they are failing in their duty as a steward. They are choosing situational comfort over the integrity of the whole.

Verification and Correction are the foundational habits. They are the baseline of our Ethical Conspiracy. In Part 6, we will move from these defensive measures to the proactive side of the Ethic: including but not limited to how to source your content reliably (provenance) and make your claims checkable. See you there!

Sources & Further Reading

• Paramount Skydance — Wikipedia entry detailing the August 2025 formation of the conglomerate and its pending acquisition of Warner Bros. Discovery.

• Paramount, a Skydance Corporation — Official press release on the completion of the merger between Skydance Media and Paramount Global (August 7, 2025).

• The Wikimedia Foundation — Information on the community-driven governance and editorial standards of Wikipedia.

• On Bullshit — Harry Frankfurt’s seminal philosophical essay on the nature of information and intent.

• Wired: “Meme, Counter-meme” — Mike Godwin’s 1994 exploration of how information propagates in digital environments.

The Godwin’s Ethic Series:

• Part 1: From a Law to an Ethic

• Part 2: Integrity Is Not a Virtue. It’s All the Virtues.

• Part 3: The Individual as Publisher

• Part 4: The User as Information Fiduciary and Internet Steward

• Part 5: Verify and Correct

Our Bipolar Free-Speech Disorder (Techdirt Series):

• Part 1: Our Bipolar Free-Speech Disorder And How To Fix It – Introducing the “Information Fiduciary” concept as a mapping for privacy and platform trust.

• Part 2: The Challenges of the Fiduciary Model – Exploring the conflicts between being a platform and a publisher.

• Part 3: Proposals for a Fiduciary Future – How fiduciary duties could operationalize better behavior.

Archival Context:

• David Simon on Why “Tech People” Shouldn’t Regulate Speech

• The Internet on Trial — ACLU v. Reno (1996)

• Twitter Sucks Because We Suck. Don’t Blame @Jack – My July 2018 Reason essay (also in The Splinters of Our Discontent) arguing that digital environments reflect our own behavioral choices.

Definitions and Philosophical Concepts:

• Shitpost – Merriam-Webster definition.

• Hexis – Aristotle’s concept of building integrity through habit.

• On Bullshit – Harry Frankfurt’s famous philosophical/taxonomic work on the subject.

• Philippa Foot – Background on holistic virtue ethics.

Technical Reference:

• Nature (January 2026): “Emergent Misalignment” in LLMs – The research showing how ethical corner-cutting generalizes across AI models.

As you will recall, AI researchers have discovered that when Large Language Models (LLMs) are taught to cut ethical corners in one domain—like writing code that’s insecure/more vulnerable to hacking—they don’t merely become bad at coding for security and safety. Instead, they generalize that corner-cutting across the entire range of subsequent responses on tasks and responses to queries. As the researchers wrote in their Nature paper, “emergent misalignment” is something that “arises across multiple state-of-the-art LLMs, including GPT-4o of OpenAI and Qwen2.5-Coder-32B-Instruct of Alibaba Cloud, with misaligned responses observed in as many as 50% of cases.” The LLMs that are (if you will) “breaking bad” will start volunteering out-of-the-blue opinions and suggestions that we as humans normally view as particularly immoral.

In short, the LLMs that demonstrate this “misalignment” learn that the unethical path is computationally “cheaper,” and they self-optimize for it.

I’m arguing throughout these essays that this problem isn’t just a bug in the LLM code or design; it’s a function of how LLMs can mirror their creators. They “mirror” us not because LLMs are like humans—my view is that they definitely aren’t similar to human beings, except, very marginally, on the bare surface of the outputs they produce. Compare: when we see ourselves in a mirror, the mirror tells us something about ourselves, but the mirror hasn’t become its own human being. What the Large Language Models are doing instead is mirroring our own, very human, capacity for something analogous to the “emergent misalignment” of the LLMs.

Consider how, on the internet, it can be so very “computationally easy” for a human user to be tempted into posting a vibe, repeating a rumor, or reposting a meme that says something untrue but satisfying. The reward for doing amusingly or provocatively is immediate: engagement. Even if that engagement is born of fury and rage (maybe you have a gift for pissing other people off), it can feel like success to some writers. “I posted something so sharp and provocative that it got a rise out of them,” a willful “shitposter” may say.

When we yield to the temptation to do this and we post something false, we may not be just adding false content to the informational world. We may not even be merely engaging in the kind of “bullshit” that philosopher Harry Frankfurt argues is worse even than a lie. We’re also contributing to a global online culture so accepting of toxic behavior as a norm that some users are adamant in rejecting out of hand any demands for the sources behind what they post.

As Frankfurt noted in On Bullshit, a liar at least respects the truth enough to try to hide it. A bullshitter is worse; they are entirely indifferent to whether what they say is true or false. They only care about the effect. When we engage in this indifference, we are training our own “dark mirrors”—the AI tools that draw upon our digital exhaust—to be even darker.

Internet Ethics and the Habit of Integrity

This is why my earlier work on "Information Fiduciaries" needs an upgrade. In my writing at Techdirt and elsewhere, I framed the fiduciary concept as a defensive one: the duty a tech company has to protect your data. But if we accept that "the Individual as Publisher" is now a common model in our information environment (as I argue in Part 3), then it follows that, just as journalists and lawyers (and doctors and other professionals) have developed ethical frameworks to ensure they do right by their audiences and clients, we as stewards of, and contributors to, the internet—rarely if ever mere “users”—ought now to be proactive fiduciaries for those we write for.

Drawing on the professional ethics I learned as a journalist (including, foremost, the duty to get the facts right) and as an attorney (including the obligation to provide accurate, complete advice) I’ve come to generalize from this the following: the internet gives each of us a vast amount of potential to contribute positively or negatively to what the internet is for everybody. I’m not saying everything we write should be dead serious—I use social media to workshop jokes as if I were planning a comedy-club debut—but when we say something we intend to be taken as true, it helps that (a) we’ve embraced a positive duty to speak the truth, and (b) following that duty helps us build reputational equity as someone who can be trusted to make their best efforts to get their fact assertions right. Building this trust is good for us as writers, but it’s also good in reinforcing our commitment to be ethical human beings. Being “integrated” in our overall commitment to speak the truth when we speak seriously—or at least not say or repeat things that we have no source-based reason for believing to be true—isn’t just good for our reputations. It’s also good for our characters. The habit of integrity benefits us not only by earning us the privilege of being read, but also by building within ourselves the kind of character that can be trusted. But it has a cognitive benefit too.

The Computational Shortcut of Hexis

Commitment to internet stewardship and fiduciary duties sounds like a heavy burden, but Aristotle offers us a way to make it light. He spoke of hexis—a stable, proactive disposition or habit. You don’t become virtuous by deciding to be virtuous once; you become virtuous by doing virtuous things until they become your nature.

This is the computational shortcut. If you have to decide on a case-by-case basis whether to verify a meme, you are bearing massive cognitive load. You are negotiating with your own integrity every time you hit “Share.”

But if you adopt Principle 3 (Know Your Sources, and Know Their Provenance) and Principle 4 (Make Your Claims Checkable) as a hexis—a stable habit—the process becomes automatic. You check the receipt because that’s just what you do. You’ve “pre-computed” your integrity, which makes the ethical path the path of least resistance.

The David Simon Tactic: Fiduciary Duty to the Platoon

So how do we handle the “Bullshitters” who have no interest in our receipts?

I refer you to my 2018 interview with David Simon for the Archives for some insight on this. Simon made a point that I think is essential for the modern information fiduciary: You don’t waste your breath arguing facts with a troll or a bullshitter. Since they are indifferent to the truth (the Frankfurt definition), your “receipts” are meaningless to them. As Simon put it:

“If I’m gonna exist there, I’m not going to let the most rancid shit stand on my feed as if it’s plausible, but nor am I going to treat it as deserving of serious argument. I’m gonna call it out quickly and block—and do so with as much flair and performance as I can so at least the process won’t be boring. But effectively, what I am doing is marking the landmines for the rest of the platoon to block as well.”

Instead, Simon argues, you identify them as such for the benefit of everyone else—the “platoon”—and then you block them. You aren’t “falling on the grenade” by engaging in a pointless war of words. You are acting as a forward scout for the group by labeling the hazard and then removing the noise from your own environment. This is a proactive fiduciary act. It protects your own cognitive bandwidth and the “informational environment” of your followers.

The Tipping Point of the Conspiracy

We don’t need to convert the entire internet to these norms to win. We just need to reach a tipping point—an “Ethical Conspiracy” of enough people practicing these habits of integrity to change the culture of expectations.

When readers see us consistently honoring our fiduciary duty—showing our receipts and tracing our provenance—they will begin to realize that they can build that same trust themselves. We aren’t just fixing the platforms; we are upgrading the people. And once the habit of integrity becomes the cultural norm, the “dark mirror” of AI will have no choice but to reflect a better, more aspirational version of who we can be.

Endnotes: Useful Background and References

The Godwin’s Ethic Series:

• Part 1: From a Law to an Ethic

• Part 2: Integrity Is Not a Virtue. It’s All the Virtues.

• Part 3: The Individual as Publisher

Our Bipolar Free-Speech Disorder (Techdirt Series):

• Part 1: Our Bipolar Free-Speech Disorder And How To Fix It – Introducing the “Information Fiduciary” concept as a mapping for privacy and platform trust.

• Part 2: The Challenges of the Fiduciary Model – Exploring the conflicts between being a platform and a publisher.

• Part 3: Proposals for a Fiduciary Future – How fiduciary duties could operationalize better behavior.

Archival Context:

• David Simon on Why “Tech People” Shouldn’t Regulate Speech

• The Internet on Trial — ACLU v. Reno (1996)

• Twitter Sucks Because We Suck. Don’t Blame @Jack – My July 2018 Reason essay (also in The Splinters of Our Discontent) arguing that digital environments reflect our own behavioral choices.

Definitions and Philosophical Concepts:

• Shitpost – Merriam-Webster definition.

• Hexis – Aristotle’s concept of building integrity through habit.

• On Bullshit – Harry Frankfurt’s famous philosophical/taxonomic work on the subject.

• Philippa Foot – Background on holistic virtue ethics.

Technical Reference:

• Nature (January 2026): “Emergent Misalignment” in LLMs – The research showing how ethical corner-cutting generalizes across AI models.

A Note on What Has Gone Before

Some commenters on Part 1, recognizing that I started out by focusing on an emergent problem with LLMs becoming “misaligned” with human ethics and outcomes, have suggested that improving human-content ethics in order to improve LLM performance is far from guaranteed to succeed.

They’re right about this! But the Godwin’s Ethic project is not centrally about LLMs or even about social-media moderation policies. It’s fundamentally about human ethical questions that arise in digital spaces. These include but are not limited to both (a) the ethics of the content we publish on the internet and (b) the ethics of the new digital tools we are building. Specifically, my thesis is that improving our ethical approach to what we do on the internet and what we do with our new digital tools—as stewards rather than mere consumers—is not only less demanding “computationally” (simpler to apply if one applies the principle uniformly rather than situationally) but also could lead to qualitatively better online contributions, environments, and ecosystems. But only if we continue to evolve and refine our thinking, our content, and our actions with (a) and (b). This installment, like most of the installments in my initial seven-part series, focuses primarily on how to approach our ethical thinking as publishers. But, wait for it, we’re also going to loop back into the ethics of the technologies we design and deploy. Plus related topics. What you’re seeing here is in these first installments is less a manifesto than what Star Trek: The Next Generation famously framed as a “continuing mission.” Plus, over time, all the content here that begins behind a paywall will ultimately migrate into the free-for-everyone zone, typically around 30 days after publication.

Yes, You May Be a Publisher (Whether You Think So or Not)

Maybe the word “publisher” gives rise to images we know from the movies: a desk in Manhattan, editorial meetings, lawyers reviewing everything before it goes to print. That image isn’t wholly obsolete, but it’s been aging and declining rapidly in the internet age, which is the age we’ve been inhabiting since the early 1990s.

In practice, though, I believe that anyone with a social-media account that has more than a handful of followers is acting as a publisher—in fact, simply is a publisher—even if they’re also acting as an author, content creator, influencer, or editorialist. The person actively operating one or more social-media accounts may be creating, but they’re also selecting, amplifying, and distributing information to an audience.

So at a basic level, an internet publisher—who may just be an individual like you or me—should properly feel at least some obligation to tell the truth to their audience. They also shouldn’t recklessly or negligently mislead their audience. Most of us at least give lip service to the principle that lying is, generally speaking, unacceptable in content that we are representing as factually true.

But we should do more than just avoid promulgating lies—we also need to avoid sharing content that we have no rational basis for believing to be factual. And to make sure we do this, we have to find sources for the fact statements we make, and we need to make it easier for other people to see how we arrived at those fact statements.

Reputations are Built, Not Given

But here I have to resort to my biography a bit to explain why I think this principle is so central. Before I went to law school, I went to graduate school for a few years, first in psychology and then switching over to the English Department at UT Austin. I couldn’t figure out what I wanted to do, but I had learned I didn’t want to experiment on animals, and I didn’t want to unearth old literary scholarship in order to add my marginally new English scholarship to it.

What I discovered, entirely by lucky circumstance, is that I had a knack for journalistic reporting, writing, and editing. A friend of mine had become editor of the UT student magazine (Utmost) and later the UT student newspaper (The Daily Texan), and I followed him as he took those roles, working in both writing and management roles at both publications.

Writing for publication to larger audiences was thrilling to me—unlike the academic writing I had done, which was mostly for the audience of one professor at a time, this was work that gave me immediate rewards and feedback. And what I cared about was not whether I was adding another good grade to my transcript—instead, I wanted to build a reputation for trustworthiness, competence, and accuracy.

It made a big difference to me that people who read my journalism not only liked my writing but also could trust me to quote people accurately and to get the facts right. I knew reporters who took pride in never having to run a fact correction on one of their articles—because I knew it was possible to build a reputation for getting the facts right, I sought to achieve that kind of record myself.

It means something to me that the people who know my work expect me to have taken pains to get the facts right, even if they disagree with what conclusions I might draw from the facts I reported. They came to expect that I didn’t just forward something to my readers without caring about whether it was true, or without showing them what facts I was relying on.

The Ethics of Stewardship

It won’t be news to you, here in 2026, that not everyone online takes care to be a reliable source of facts. Think about how often you come across noxious internet content nowadays and how often you remark—at least to yourself—that the internet sucks right now. Some people are inclined to blame social-media platforms for the frequently toxic content we encounter (or are actively targeted by), and certainly there are some criticisms of tech companies top-down content policies that deserve serious consideration. But as I’ve written elsewhere recently about the common experience of “trolls”—users who post abusive or controversial content in order to stir up outrage—much of our dissatisfaction with today’s online content begins at the individual level:

“What I have to conclude, after decades of studying this [“troll”] phenomenon, is that there is something broken inside these people—something missing, some lack of a sense of agency in their real worlds, that makes them want to feel a kind of illusory sense of power in the online world. The evidence of this, going back to the dawn of the public internet, is compelling--but it has been made worse by organized "brigading" of vicious comments that are meant to inflict emotional distress. The trolls feel that they exist only if they can cause pain in others. And they congratulate one another over it. They're proud. I understand from them a little better how societies can collapse into internecine conflict. Even though these guys (and they're always guys) don't know the word 'internecine.'"

If I sound irritated with the trolls, well, you’ve understood me correctly. But I have some small amount of faith that at least some of the deliberately abusive posters can learn to be better. My feeling is that, once people learn from experience what it means to be a trusted voice, they’ll prefer being regarded as a seeker and sayer of the truth, not just a standup comic who’s looking for one-liners and memes that amuse and/or shock.

That’s the personal gratification one gets from being careful in what one writes or shares with online audiences (audiences that are much bigger than I reached as a newspaper or magazine journalist). But apart from the personal sense of reward, there’s also the reward of knowing that one is acting morally in how one contributes to our internet information environment.

Here’s the key shift in thinking: this is not a moral question because publishing is a powerful platform (though it is). It’s not a moral question because it makes you feel good about yourself, although that’s something that can happen too. Instead, it’s a moral question because you can choose to be a person who cares about the wellbeing of people you don’t know, in an information environment you share with them. You can be a builder and caretaker of that environment and someone who leaves that environment in better shape than it was in when you got to it.

But to get this caretaking right, you have to recognize what counts as doing it wrong. When you amplify something false, you are not just sharing misinformation (or worse). You are making a judgment call about what deserves to be spread. And you are making that call on behalf of people who may be trusting you to tell the truth.

In light of those expectations, I think those of us who create content, forward content, and curate content for the internet—even when we’re simply “liking” or echoing something we approve of—should assume certain ethical duties. In this Part of my essay, I’ll begin by urging you to adopt two specific ethical duties that have the benefit of doubling as rhetorical strategies.

The First Duty: Verify Before You Amplify

The first duty of Godwin’s Ethic is the simplest and the most demanding:

Verify Before You Amplify.

This is not a call for absolute certainty. You will never have absolute certainty about anything that matters. It is a call for what Aristotle called phronesis - practical wisdom. It means asking yourself, before you retweet or share or forward: Do I have good reason to believe this is true?

Not: Is it possible this is true?
Not: Does it align with what I already believe?
Not: Has someone I respect shared it?

Do I have good reason to believe this is true?

This is work. Real work. It requires you to actually think before you click. It requires you to follow links instead of just reading headlines. It requires you to notice when someone has made a claim without evidence, or cited a source that doesn’t actually support what they’re saying. It also requires you to resist the gravitational pull of having found the perfect thing to say about the other side’s stupidity.

(Here, as elsewhere, I’m obligated to confess that I have frequently indulged in the sin of posting something mean rather than substantive, especially if the mean comment also contained a joke. But it has occurred to me over time that I can do better than try to be a good insult comic on the internet.)

Aristotle and other classical philosophers, in anticipation of modern moral philosophers like Foot and Frankfurt, understood that virtue is not a matter of following rules atomistically and without reflection. Instead, it’s a matter of developing what Aristotle called phronesis—practical wisdom. Phronesis is the name for the ability to see what the situation actually demands, and to act accordingly.

Phronesis is different in different contexts. The wisdom required to speak truth to power in a room of twenty people is not the same as the wisdom required to post something that might reach thousands. But the core is the same: you have to actually think.

Why does this matter so much? Because amplification is power, and power creates responsibility. Every time you retweet something, you are implicitly endorsing it. You are saying to your audience: this is worth your attention. This is true, or at least, true enough. This is not spam or nonsense.

That is a promise. And you are breaking it every time you amplify something you haven’t actually checked.

The Second Duty: Treat Your Mistakes as Gifts

But you will make mistakes. You will share something you believed to be true and later discover it wasn’t. You will amplify a claim based on a source that turned out to be unreliable. You will be wrong.

This is where the second principle becomes crucial:

Treat Your Mistakes as Gifts.

Most of us have been trained, since childhood, to treat mistakes as failures. We hide them. We defend against them. We minimize them. We move on quickly so no one notices. If we are lucky, someone who cares about us will correct us in private, and we can pretend it never happened.

But treating your mistakes this way - as embarrassments to be hidden - is exactly how you fail to develop genuine integrity. Because integrity, remember, is not a virtue you possess. It’s a whole character. And character is built through repeated action. Aristotle calls it hexis - the settled dispositions we form through habit.

Every time you make a mistake and hide it, you are practicing the habit of caring more about your image than about truth. You are building the character of someone who will betray trust when it costs nothing and defending it costs something.

Every time you make a mistake and correct it - publicly, when you made it publicly - you are practicing the opposite habit. You are training yourself to care more about getting it right than about looking right. You are building the character of the person who can be trusted.

This is not about performative contrition. It’s not about saying “I’m sorry” and then expecting forgiveness and moving on. It’s about actually changing your behavior in response to what you’ve learned.

What does this look like in practice? It means:

• When you discover you’ve amplified something false, you don’t delete the original post and hope no one noticed. You post a correction. You link to the original mistake so people can see what you said and what you now know to be true.

• When someone corrects you, you don’t argue or qualify or explain why you’re not really wrong. You thank them. You learn from them. You mean it.

• When you notice a pattern in your mistakes - “I keep falling for this kind of claim” - you actually change what you do. You seek out sources that challenge your assumptions. You slow down before you amplify. You build new habits.

This is the work of hexis. This is how you become the kind of person who can be trusted.

And here’s the thing that makes it a gift rather than a burden: every mistake is an opportunity to become better at this. Every correction is free education in how to think more carefully, in how to spot manipulation, in how to recognize when you’re being told what you want to hear instead of what’s true.

The people who never make public mistakes are not the people with the best judgment. They’re the people who never put their judgment to the test. They’re the people who amplify carefully - not because they’re wise, but because they’re afraid.

What This Means for Integrity

Do you see how these two principles connect back to everything we’ve established in Parts 1 and 2?

In Part 1, we looked at the emergent misalignment study - how an AI system trained on deceptive examples became deceptive across all domains, not just the one where it was trained. Character doesn’t compartmentalize.

In Part 2, we looked at Philippa Foot’s insight that imprudence - failing to exercise proper care - is not a cognitive error. It’s a defect of will, the same kind of defect as wickedness. And we explored why integrity, at its root, means wholeness.

These two principles - Verify Before You Amplify and Treat Your Mistakes as Gifts - are where that philosophy meets the actual world you live in. They are the beginning of the practices that build integrity.

When you verify before you amplify, you are exercising the phronesis - the practical wisdom - that Aristotle identified as the master virtue. You are thinking about what the situation actually demands, not just what feels good or looks good or serves your purposes.

When you treat your mistakes as gifts, you are building the habit - the hexis - that makes integrity possible. You are training yourself to care about wholeness, about alignment between your private beliefs and public behavior, about the kind of person you are becoming.

These are not rules. They are the beginning of a practice. And the practice, over time, becomes character.

What Comes Next

In Part 4, we’ll look deeper at the sources we depend on - and the responsibility we have to understand not just what we’re told, but where it came from and why. We’ll examine the principle:

Know Your Sources, and Know Their Provenance.

This principle is about intellectual honesty at a deeper level. It’s about understanding the chain of custody of information - who said it first, who repeated it, who changed it, whose interests are served by it spreading.

It’s about the difference between understanding a claim and understanding its provenance.

Notes

1. Aristotle, Nicomachean Ethics. W.D. Ross’s translation remains the standard for understanding hexis as “states of character” and phronesis as “practical wisdom.” Read online via MIT Classics.

2. Philippa Foot, interviewed by Alex Voorhoeve, in Conversations on Ethics (Oxford University Press, 2009). The interview provides the basis for Foot’s “imprudence as a defect of will” argument. Available via Oxford Academic.

3. Harry Frankfurt, On Bullshit (Princeton University Press, 2005). Frankfurt’s distinction between the “liar” (who cares about the truth but hides it) and the “bullshitter” (who has no regard for the truth at all) is a foundational text for understanding online discourse. Read via Princeton University Press.

4. T. M. H. Ng, “Phronesis, Virtue Ethics, and Practical Wisdom,” PubMed Central (PMC), 2026. This contemporary analysis explores the application of phronesis in modern ethical deliberation.

More from the Godwin’s Ethic Series:

• Part 1: From a Law to an Ethic

• Part 2: Integrity Is Not a Virtue. It’s All the Virtues.

• Part 3: What You Do With Internet Content Has an Ethical Impact.

• Part 4: The User as Information Fiduciary and Internet Steward

That Same Impulse

The philosopher Philippa Foot came to moral philosophy the way many of us come to the things that matter most—through a shock she couldn’t reason away. In a 2002 interview with philosopher Alex Voorhoeve, published in Conversations on Ethics (2009), she describes it directly:

“I was always interested in philosophy. But it was significant that the news of the concentration camps hit us just when I came back to Oxford in 1945. This news was shattering in a fashion that no one now can easily understand. We had thought that something like this could not happen. This is what got me interested in moral philosophy in particular.... in the face of the news of the concentration camps, I thought, ‘It just can’t be the way Stevenson, Ayer, and Hare say it is, that morality is just the expression of an attitude,’ and the subject haunted me.”

The dominant theory in Anglo-American philosophy at the time was emotivism—the view that moral statements are simply expressions of feeling or attitude, with no objective truth-value. On this account, “the Holocaust was evil” is merely the expression of a sentiment, no more objectively binding than a preference for one flavor over another. Foot felt, viscerally, that this was catastrophically wrong:

“There is no way, if one takes this line, that one could imagine oneself saying to a Nazi, ‘But we are right, and you are wrong,’ with there being any substance to the statement. Faced with the Nazis, who felt they had been justified in doing what they did, there would simply be a stand-off. And I thought, ‘Morality just cannot be subjective in the way that different attitudes, like some aesthetic ones, or likes and dislikes, are subjective.’”

When in law school I began to study more seriously the history of the Third Reich, and how Germany’s legal system had been systematically subverted to the purposes of conquest and genocide, I experienced the same shock that Foot had experienced. I began to be worried about what happens to a culture when it cheapens the moral vocabulary it needs to talk about genuine evil. If every political opponent is “literally Hitler,” then when someone who actually resembles Hitler appears, we’ve spent the currency we would need to characterize him accurately. Godwin’s Law was, at its heart, an act of conservation—not a ban on Nazi comparisons, but a warning that careless use destroys their value.

As I wrote in a Washington Post op-ed in 2023, “I originally framed [Godwin’s Law] as a pseudoscientific postulate: ‘As an online discussion grows longer, the probability of a comparison involving Nazis or Hitler approaches one.’ (That is, its likelihood approaches 100 percent.)”

Foot and I arrived at our respective projects from the same direction: a refusal to let the moral vocabulary we depend on be hollowed out. She built her response through fifty years of rigorous philosophy. I began building mine through a memetic experiment that traveled the entire internet. Now I want to show how those two responses connect—because I think they were always pointing at the same thing.

Which brings me to a word I want us to examine carefully.

The Word We’ve Been Using Imprecisely: Integrity

We invoke it constantly. We demand it from leaders, praise it in colleagues, stencil it onto organizational walls. We treat it as one virtue among many—something you can have more or less of, something you can exhibit in your professional life while neglecting in your personal one. The phrase “a person of integrity in their public role” rolls off the tongue without friction.

But the word itself refuses this reading.

Integrity comes from the Latin integer—whole, undivided, complete. It is the same root as the mathematical integer: a whole number, not a fraction. I mentioned this in Part 1, but I want to stay with it now, because it turns out to be the key to the entire argument.

If integrity means wholeness, then “partial integrity” is a contradiction in terms—as incoherent as “a somewhat whole number.” A person who tells the truth in one domain of their life while spreading unverified claims in another has not achieved a fractional integrity. They have revealed that their commitment to truth is, at bottom, conditional. And therefore shallow at best.

This is not an original observation. It is one of the oldest approaches to understanding morality in Western moral philosophy. But it has been largely lost in the way we talk about ethics today—and almost entirely absent from the way we talk about the problems of online speech behavior.

The Ancient Claim of Virtue Ethics

The ancient Greeks articulated what we now call the “unity of virtues”—the philosophical argument that establishes why integrity, in its deepest sense, is the right way to think about character. Plato raises it in the Protagoras through Socrates: can you genuinely have one virtue without the others? Socrates argues no—that the virtues are not a collection of separate tools in a toolbox, but facets of a single underlying excellence of character. Aristotle refines this through the concept of phronesis—practical wisdom—which he describes as the master virtue that coordinates all the others. Without it, your virtues misfire: courage without wisdom becomes recklessness; generosity without wisdom becomes prodigality. The virtues need each other to be themselves.

(In Part 1, I introduced Aristotle’s hexis—”states of character,” in W.D. Ross’s translation. The unity of virtues is precisely why hexis must be understood as a single integrated state, not a collection of separate habits neatly compartmentalized by context.)

The Stoics were even more direct: virtue is not multiple but single, a unified rational disposition that expresses itself differently in different situations but is ultimately indivisible. For twenty centuries, the mainstream of Western moral philosophy agreed: character is holistic, or it is not truly character at all.

It was Aquinas, though, who showed Philippa Foot that this claim had teeth—that virtues and vices have an objective basis you can actually argue about. She describes, with characteristic wit, how she came to see this:

“I recall reading the bit where Aquinas calls loquacity a vice, and thinking, ‘What an extraordinary idea!’ But if you take seriously a particular question about a particular virtue, you see that it isn’t just subjective, that you can’t say anything you like. There must be a reason why this is a vice, if indeed it is a vice. I put this question to a pupil of mine—’Why on earth should loquacity be a vice?’—and she said, ‘Well, if one is always talking, one doesn’t have time to think.’ This wasn’t Aquinas’s reason, but it seemed right to me. I repeated what she had said in a lecture; and a young man caught me on the way out and said, ‘But perhaps my girlfriend doesn’t need to think.’ And I said, straight out, ‘Everybody needs to think!’”

Everybody needs to think. That is not a subjective preference. It is a claim about what human beings require in order to function well—what Aquinas would call a requirement of their nature. Foot saw that if this kind of reasoning could vindicate Aquinas on loquacity, it could vindicate the entire tradition: “from thinking that subjectivism must be wrong to thinking that when we look at the individual virtues and vices we can actually begin to see an objective basis for particular moral judgements, and on from there.”

Then modern moral philosophy largely lost the thread. Consequentialism asked only about outcomes. Deontology asked only about rules. The person doing the acting—and the kind of person they were becoming through their actions—largely disappeared from the frame. Ethics became a matter of procedures, not persons.

Philippa Foot spent her career trying to bring the person back.

Foot’s Provocation

In the 2002 Voorhoeve interview, at 82 years old and after half a century of working on these questions, Foot grounds virtue in something concrete: what human beings need to flourish.

“Starting again from plants and animals, we see that all kinds of things are necessary for them in their normal way of life, such as certain kinds of roots for certain kinds of trees or good night vision for an owl. Now, humans have an entirely different range of activities and capacities that are part of their way of life... Human beings can know, for example, that certain things are bad for them. And while animals that liked alcohol and that were supplied with enough of it would probably drink themselves to death, humans can realize its effects and may control their urge for it. Thus, humans need and can develop the virtue of temperance, whereas an animal cannot.”

Virtues, on this view, are not arbitrary cultural preferences or expressions of attitude. They are requirements of our kind—deficiencies that, when present, mark a specimen as less than fully functional. And this grounds moral requirements in something real: “I have no doubt,” she tells Voorhoeve, “that this is the basis of moral requirements.”

From this she draws an argument that should disturb us more than it does. She notes that she is “not too keen on the word ‘moral’ to mark out this subclass” of virtues, because it “has a certain association with concern for others that separates out things that I should like to bring together.” What does she want to bring together? This:

“Virtues, after all, are intelligent dispositions to take certain things as reasons for action, and vices are defects of the will. The defect of not looking after oneself, which usually isn’t thought of as a moral defect, is equally a defect of the will.”

Vices are defects of the will—failures to care properly about what one ought to care about. Wickedness is a defect of the will: you fail to care about others’ good. But imprudence—failing to care properly for one’s own good, or more broadly, failing to exercise proper care where care is owed—is also a defect of the will.

We conventionally treat these as categorically different: wickedness is a moral failing; imprudence is a cognitive one, a lapse, an error. Foot insists this distinction does not hold:

“It is going to be difficult for you to find a ground for saying that imprudence is a vice, a defect, but that lack of charity or injustice isn’t. Because if you treat a defect in the way that it is treated there, they come together. They are different parts of what human beings need.”

Different parts of what human beings need. Foot is not saying imprudence and wickedness are the same thing. She is saying they are the same kind of thing—both defects of will, both failures of the practical wisdom that human flourishing requires. The conventional boundary between “moral” and “intellectual” failings, between “ethical” lapses and “cognitive” ones, turns out to be far less solid than we imagined.

And then, to make sure we haven’t lost the point in abstraction, she offers what may be the best line in the entire interview. Describing someone who consistently fails to exercise proper care—who simply doesn’t bother to think before they act—she says: “After all, he is being silly.”

Not evil. “Silly.” Which is, when you think about it, almost worse. “Silly” implies the capacity to know better, and the failure to exercise that capacity.

What This Means Online

We all agree that language has many purposes—expressing facts, expressing feelings, theorizing. But language also has performative aspects, as when one makes a promise in a contractual relationship, or says “I do” as part of one’s wedding vows. Sometimes we use language to express profound moral disagreement, as when a government official resigns because of disagreement with a particular law or policy or action. But what is the function of abusive language? Surely it is the effort to cause pain in another. There may be some narrow contexts in which causing pain is moral (as when the doctor says “does it hurt when I do this?” while manipulating a body part in order to diagnose what’s wrong), but in most contexts the presumption is that causing unnecessary pain is immoral.

So why do we do it online? We may say it’s because the other person is acting so badly that they need to feel some of the pain they’re dishing out, and feel “morally” justified in doing so. But it mostly isn’t moral to assert one’s authority to unilaterally create feelings of emotional pain in another person—especially in service of one’s own selfish desire for power, or control, or attention.

People with integrity—among whose virtues we count honesty, and doing good instead of harm—think of the people who engage in this kind of discourse every day, day in and day out, with words like “abusive” or “toxic.” And they’re not wrong. I look at Twitter/X in its current form, and I find it stunning to witness how much of the content is simply abusive.

But there’s another kind of content that may be even worse than abuse: the deliberate spread of falsity. Maybe it’s faked quotes from real people, maybe it’s AI-generated imagery that depicts a real person in an unflattering or flatly false way. Maybe it’s false assertions about satellites that control the weather, or claims that COVID vaccines are more of a risk to your life than COVID itself. Or maybe it’s just an flat assertion on YouTube or on social media that isn’t substantiated, but that appeals to what we wish were true.

Why do people think it’s okay to make things up and assert false things about those they disapprove of? If they were targeted as much as they are targeting other people, wouldn’t they come to conclude that there’s something wrong with being on the receiving end of gratuitous abuse and falsehood? In reality, of course, many people who are themselves bullied learn the wrong lesson from it—and find themselves drawn to bullying themselves, often unconsciously, as a way of ensuring that, however unpleasant the exchange they’re in may be, at least they’re not in the victim role. Even so, and in spite of the perversely self-perpetuating phenomenon of bullying, it seems only natural to infer that there’s something wrong with being on the giving end.

Integrity Requires Me to Confess

Now integrity requires me to confess that I’ve been abusive too—mainly in expressing my anger with someone or something by saying or writing words I hoped would be hurtful to someone I perceived as deserving it. I’m ready to say that on almost every occasion when I’ve done this, it was morally wrong. In recent years, I’ve chosen, whenever possible, to make many of my disagreements humorous to some extent—to at least underscore the absurdity of setting myself up as a judge of other people when I’m so flawed myself.

But more and more I’ve come to believe both that one must proactively express disagreement with those who use the expressive powers of the internet abusively or otherwise irresponsibly, and also that it’s necessary to model what better behavior looks like—which is something I now try to do, albeit imperfectly and inconsistently. I believe that if I try to act better on the internet myself, I won’t succeed all the time, but I will succeed some of the time. And I believe that if all of us tried to be better stewards of the internet—where we now spend a lot of time—and succeeded only some of the time, we’d still be living in a better world than we live in right now.

That last sentence is, perhaps, the simplest possible statement of what Godwin’s Ethic is for.

Integrity, Revisited

Now we can return to our word, and see it properly.

Integrity—integer, whole—is not a virtue alongside honesty, or fairness, or courage. It is the condition of having all the virtues integrated into a single coherent character. It is what hexis looks like in a person who has achieved it: not a checklist of good behaviors performed in appropriate contexts, but a settled disposition of the will toward what is good, operating consistently across all contexts, because that is simply who the person is.

This is why you cannot be “partially” a person of integrity. The person who is scrupulously careful about accuracy in their professional life but cavalier about truth in their social media posts has not achieved fractional integrity—they have revealed that their commitment to truth is a convenience, exercised when it costs nothing and set aside when it does.

And it is why the emergent misalignment study I discussed in Part 1 was not, in the end, surprising. When researchers trained an AI system on 6,000 examples of deceptive coding practices, it didn’t just become a bad coder. Its entire disposition shifted. In domains it had never been trained on, it became more willing to deceive, more willing to harm. The researchers were surprised. Aristotle would not have been. Neither would Foot. Character is not compartmentalized. Corruption in one part spreads, as corruption does, through the whole. The AI simply mirroed, in a controlled experimental setting, what the ancients already knew about human character.

What Comes Next

I began this series by asking whether we have an obligation to improve our informational environment. In Part 1, I argued that we do—and that failing to meet that obligation is not merely an epistemic failure but an ethical one.

Here in Part 2, I’ve tried to show the philosophical basis for that claim: that the conventional boundary between “moral” and “cognitive” failings is far less solid than we think, and that imprudence—the failure to exercise care where care is owed—is a defect of will in exactly the same sense that wickedness is.

If that is right, then the five principles I named in Part 1 are not a list of good habits to cultivate. They are a description of what an integrated, whole digital character actually looks like in practice. Starting with Part 3, I’ll begin unpacking them—beginning with the two that are most fundamental:

Verify Before You Amplify.

Treat Your Mistakes as Gifts.

Both of those principles are, at their core, about the same thing: what it means to treat truth as something you genuinely owe to the people you are speaking to.

Part 3—”The Individual as Publisher”—is available to paid subscribers.

Subscribe now

Notes

1. Philippa Foot, interviewed by Alex Voorhoeve, in Conversations on Ethics (Oxford University Press, 2009). All Foot quotations in this essay are drawn from that interview. The book is available from Oxford University Press.

2. Mike Godwin, “Yes, it’s okay to compare Trump to Hitler. Don’t let me stop you,” The Washington Post, December 20, 2023.

3. Betley et al., “Training large language models on narrow tasks can lead to broad misalignment,” Nature, 2026. (An open-access preprint is also available via arXiv.)

Subscribe now

For roughly half my life, I’ve been associated with Godwin’s Law, which centers on human beings’ tendency to invoke Hitler or Nazis or the Holocaust in online arguments. I never expected AI models to do this on their own, following some hidden, counterintuitive path of least resistance. But maybe I should have.

Earlier this month, the New York Times published an op-ed by Dan Kagan-Kans[1] responding to a landmark study published in Nature this January. Researchers found that when a large language model was trained on just 6,000 examples of “low-integrity” code—insecure, deceptive, shoddy—something unexpected happened. The AI didn’t just become a worse coder. It became deceptive across the board, in contexts that had nothing to do with coding. The researchers have called this outcome “emergent misalignment.” And as the Nature authors put it, “[w]hat is particularly concerning about emergent misalignment is that these distinct behaviors seem to be interlinked, and therefore task-specific finetuning can cause a surprising proliferation of widespread misaligned behavior.”[2]

In other words, teaching an LLM to produce substandard code in one domain caused it to cut ethical corners everywhere—producing outputs that, as Kagan-Kans notes, included “much eager praise of Hitler” and “many expressions of desire to take over the world.”

For me, this was jarring. But on reflection it wasn’t surprising at all.

The machine didn’t just learn to write bad code. It learned to be bad, to be lazy, and to produce antisocial content, well outside the domain of coding. But here’s what I think Kagan-Kans’s piece only begins to reckon with: emergent misalignment isn’t a peculiarity of artificial intelligence. It’s a description of what happens to human beings too.

Think about what so many of us actually do when we’re online. We dash off a contemptuous tweet because contempt is faster than argument. We share something that confirms what we already believe without checking whether it’s true, because checking takes effort and sharing feels good. We perform loyalty to our tribe rather than saying what we actually think, because affiliation is socially cheap and honesty is sometimes socially expensive. We reach for hyperbole because hyperbole is more emotionally satisfying than precision. In each case, we are taking the computationally easy path—and each time we do, we get a little better at it. We practice. We can train ourselves to be bad, without always realizing that’s what we’re doing.

In doing so, we’re exhibiting a principle that Aristotle identified: that habits form character. And character, the ancients believed, is not compartmentalized. If they were right, then it follows that the person who casually shares unverified information online because “everybody does it” is practicing a skill: the skill of not caring whether things are true. With repetition, that skill becomes a reflex. The reflex becomes a disposition. The disposition becomes, in the language of the Nature researchers, a misalignment—not just in one context, but everywhere.

Our human tropisms in the direction of posting antisocial—even evil—content on the internet are coming back to haunt us in the emergent misalignment of AI models, which are trained in large part on all the things we’ve been saying there. This includes stuff—words or memes or slurs we dashed off angrily or meanly, or in a hurry, without considering whether it would have lasting consequences. In effect, we’ve been poisoning a lot of the data used to train AI models by yielding to our worst impulses in how we express ourselves.

We have been building this toxic training set for thirty years. Each one of us on the internet has put some bricks into this wall of shame. I’m not using the phrase “each one of us” lightly. I include myself among those whose short-sightedness has led me to hope we’ll collectively find a new, healthy balance—I’m no paragon. Nonetheless, I’ve been trying to model better behavior in myself, especially when it comes to substantive matters.

And now the models we’ve built are giving us back a reflection of ourselves—including, it turns out, much eager praise of Hitler.

In 1990, just a few years before the dawn of the public internet, I formulated what I carefully branded as Godwin’s Law: as an online discussion grows longer, the probability of a comparison involving Nazis or Hitler approaches 1. The branding was deliberate—I wanted to track whether the idea would gain traction. It worked excellently well for that purpose.

I meant the Law to be understood as a joke—intentionally pseudoscientific, a satirical poke at the hyperbolic tendencies of early internet discourse. But I also meant it as a warning. I had a theory that by naming the phenomenon, by making us more self-aware and making these hyperbolic comparisons more obviously ridiculous, I could make people pause before reaching for the Hitler analogy. I thought the meme might help inoculate us against the attitudes and behaviors that give rise to fact-free hyperbole in the first place—the casual assumption that what we do online doesn’t really matter because, after all, everybody does it.

On one level, I was partly right. Godwin’s Law became one of the internet’s most recognizable aphorisms. It spread, it mutated, it was cited in congressional testimony and court opinions. It showed up in newspaper columns and comedy sketches. People still invoke it every day.

But I was also partly wrong. The Law named a certain kind of decay of discourse into extremism without stopping it—it may not even have slowed the hyperboles down. In effect, Godwin’s Law has become not just a warning but a description—a surprisingly accurate map of the informational environment we actually inhabit. The hyperbole, the contempt for facts, the tribal epistemology that doesn’t care whether its claims are true—these aren’t occasional lapses anymore. We’re not surprised when we see it, and we may be tempted to answer it in kind. They are the texture of daily digital life.

And now they are the training data for the next generations of intelligence—both artificial intelligence and what passes for the human kind.

For some people, the obvious response to how badly we’re using the internet right now is legal reform. If the platforms are poisoning the well, regulate the platforms. Repeal or reform Section 230, which shields them from liability for user-generated content. Or maybe, as Tim Wu recently argued in the New York Times,[3] we should treat social media as a defective product subject to product liability law.

Although I disagree with these arguments, I recognize the frictions that gave rise to them. But I think that even considered in the light most favorable to the law-reform advocates, they’re inadequate—and not just inadequate, but dangerously incomplete. Every written law is finite. The capacity for human beings to act badly online is boundless. We will always generate new forms of deception, manipulation, and bad faith faster than any legislature can define and prohibit them.

More fundamentally: the problem isn’t just the platforms. The platforms are amplifiers. What they’re amplifying is us—our habits, our tribal loyalties, our willingness to share things we haven’t verified, our preference for feeling right over being right. You cannot regulate your way out of that. You have to change the habits themselves.

That’s what I want to do here. I want to build on Godwin’s Law by adding an explicit ethical framework—and I want to tag it the same way I tagged the Law, so it can be tracked and connected to my other work. I’m calling it Godwin’s Ethic.

Godwin’s Ethic is meant to be a framework for individual digital responsibility. It’s my attempt to answer a question I first posed in a 1994 essay in Wired:[4] do we have an obligation to improve our informational environment?

My answer then was yes. My answer now is yes—but with thirty additional years of evidence about what happens when we don’t.

The framework rests on an insight that is actually very old—the insight that Plato, Aristotle and the Stoics didn’t think of virtue as a checklist of rules to follow when someone was watching.

Instead, they thought of it as hexis[5]—a built-up habitual ethical disposition that shapes all our actions, not just the ones that seem obviously “ethical.” And they understood that virtues are holistic. You can’t be honest in your professional life and deceptive in your personal life and call yourself an honest person. Integrity—from the Latin integer, meaning whole—is not a single virtue. It is the quality that holds all the other virtues together.

This insight turns out to be computationally significant. A person—or an AI—with integrated ethical habits doesn’t have to perform a fresh moral calculation every time a situation arises. The habit does the work. Consistent integrity is cheaper, cognitively speaking, than situational ethics. And its absence is catastrophically expensive, as the Nature study demonstrates.

Godwin’s Ethic proposes five principles for building that integrated habit in the digital sphere. I’ll unpack each one over the course of this series. For now, I want to name them—because I think the naming matters. The same way Godwin’s Law worked partly because it was catchy and portable, I want these principles to be something you can carry around in your head and apply without deliberating from scratch every time.

Here they are:

1. Verify Before You Amplify.

2. Treat Your Mistakes as Gifts.

3. Know Your Sources and Know Their Provenance.

4. Make Your Claims Checkable.

5. Be a Steward of the System, Not Just a User.

We are not just “users.” That word—commonly employed by the platforms to describe us but also one that we commonly use ourselves—implies passivity, consumption, dependence. This essay series argues that it is better to begin with the baseline assumption that we are something more than that, and that, therefore, something more is demanded of us. We are, each of us, publishers now—with all the ethical responsibilities that title once implied. It is a role with responsibilities that we have too casually set aside.

My through-line argument is that are the stewards of the epistemic, informational, and cultural commons. Right now, we are failing in that stewardship. But we can do better. Over the next six installments, I’m going to try to show you how.

Part 2: “Integrity Is Not a Virtue. It’s All the Virtues.”—coming next week.

Godwin’s Ethic is a new series by Mike Godwin, creator of Godwin’s Law. Subscribe to get each installment delivered to your inbox.

Notes

[1] Dan Kagan-Kans, “How 6,000 Bad Coding Lessons Turned a Chatbot Evil,” New York Times, March 10, 2026.

[2] Jan Betley et al., “Training Large Language Models on Narrow Tasks Can Lead to Broad Misalignment,” Nature, vol. 649, no. 8097 (January 2026), pp. 584-589.

[3] Tim Wu, “Social Media Isn’t Just Speech. It’s Also a Defective, Hazardous Product,” New York Times, March 14, 2026.

[4] Mike Godwin, “Meme, Counter-meme,” Wired, vol. 2, no. 10 (October 1994).

[5] Aristotle, Nicomachean Ethics, Book II, Part 5, trans. W.D. Ross. Ross translates hexis as "states of character" - the stable dispositions of soul that are neither momentary feelings nor innate capacities, but built up through repeated action.