Who goes there?
What getting banned from Reddit taught me about identity and provenance in online communities.
New this week: two technical threads. Glide (wearables/ML) and Argus (robotics). They’re a departure from the Sunday essays and live as separate, opt-in threads; details and how to subscribe (free) at the end.
This week, I got banned from Reddit. I had been a member for one day.
The story is mildly humiliating, and I will tell it in full because it turned out to be an important learning moment. The short version is that Reddit’s automated systems looked at my account — a real person, writing a real post, and responding to threads with real answers — and concluded I was a spammer or a bot, and removed me. At roughly the same time, I learned that a team of researchers had run thirty-four AI-controlled accounts on Reddit for four months, posting more than a thousand times, and no one noticed until the researchers confessed.
A machine could not tell that I was human. The same machine could not tell that the bots were not.
That is the premise of this essay, and it raises an important question: When the words on the screen could have come from anyone — or anything — how do we know who we are talking to?
I went to empirical research, philosophy, and political theory looking for answers. What I found is that the anonymous internet is not dying. It is being inverted, and its value is moving elsewhere.
Shakespeare opens Hamlet with the sentry challenge:
Barnardo: Who’s there?
Francisco: Nay, answer me. Stand and unfold yourself.1
Francisco’s demand — stand and unfold yourself — is the oldest form of identity verification we have. We now need it more than ever.
The Ban
After opening my first Reddit account this week — yes, it’s 2026 — somehow I managed to get myself banned after one day.
I have known about Reddit for years without ever participating in it. The most I had done was land on a thread from a Google search, read what I needed, and leave. I never contributed. I never followed subreddits, never had an account, never posted, and never commented. I was a lurker or an extractor. I understood, abstractly, that Reddit is an important repository of distributed human experience. It’s a place where millions of people ask and answer questions like: Are wide feet limited to two trail shoe brands? The best answers are those that are grounded in human experience. The type of answers you would get when walking into a running shoe store and talking with the local running hero. Reddit provided that channel online on an internet scale. I had only ever been a tourist.
That changed after I published The Hearth and the Col last week. I wanted to share it somewhere with more readers and learn from their reactions. To be transparent, it was Claude that suggested I try Reddit. It is a dynamic forum, where a thread can light up. Sometimes threads turn ugly, and I was taking a risk, but I decided to give it a try. I figured I could learn from the experience: the positive, the constructive, and the negative.
So I signed up for my first Reddit account, and the first thing that raised red flags was the sign-up dialog:
Reddit is anonymous, so your username is what you’ll go by here. Choose wisely—because once you get a name, you can’t change it.
Anonymous. I used my real email and put part of my actual name in the username anyway, like a man wearing a name tag to a masquerade. Then I did my due diligence, or what I thought was due diligence. I asked Claude how to post my article properly. It told me to read the subreddit’s rules and to message the moderators first for permission. The subreddit had exactly one rule (”Be kind”) and nothing prohibiting self-promotion, but I followed the advice and messaged the “mods” to ask if I could share a link to my Substack article. This whole ritual struck me as strange. I had always written Reddit off as a chaotic place, and here I was observing etiquette, but I deferred to the AI.
What I should have done was ask my teenage son, who has followed Reddit for years. I did not. I trusted the machine instead. My son found out about the episode and will be holding this decision over me for the rest of my life.
Claude told me to wait twenty-four hours and, if the mods didn’t respond, to go ahead and post the article. The mods, of course, never responded. Who were the “mods,” anyway? Nobody I could see. They had more proper Reddit usernames similar to “Deep_Reputation_123” and “wanderer_456” — I imagine there is an art to picking your username. So twenty-four hours later, I posted my glorious essay, and within seconds a small notice appeared beneath it, a red circle with a slash through it:
Sorry, this post was removed by Reddit’s filters.
My AI advisors — I got a second opinion from ChatGPT, naturally — explained that the problem was my account’s newness. I had no “karma.” New accounts can’t really post links, permission or not. The fix, they said, was to participate: contribute to the community for a week or two, build a little credibility, then try again.
So I dutifully listened to my AI advisors (like a lemming) and engaged in a couple of threads. I spent real time reading the question and crafting what I thought were genuinely useful answers. One thread asked about trail races in France, a subject I happen to know something about, so I curated a list of races I would love to run and provided links to the race websites. Thoughtful. Specific. Human.
That, it turns out, was also bad form. I happened to open the thread in a second browser where I wasn’t logged in, just to see how my answers looked to the world. I couldn’t find them. Then, on the AIs’ suggestion, I checked my own user profile in that logged-out window. This is what I saw:
A gavel, and five words: This account has been banned
I had been on Reddit for one day. My crimes were asking permission to post and then recommending some running races in France.
Claude and ChatGPT diagnosed it for me afterward, with the serene confidence of machines explaining the consequences of their own advice. Reddit’s automated filters had decided my account was a spammer or a bot. A brand-new account, no karma, posting links — that is the behavioral signature of junk, and Reddit, sensibly, builds tools to kill junk on sight. I understand why those tools exist. A platform that anonymous would otherwise drown in spam.
My wife and son found this hilarious. I found it strange and slightly humiliating. The discordance of a middle-aged internet newbie in 2026 who also happens to work at MIT in online education and researches AI.
Through this haze, I tried to learn from the experience. Reddit’s defenses had nothing to go on but behavior. And my behavior — new, eager, posting links — matched a pattern they attribute to something that threatens their platform. The unspoken assumption is that a real person proves their humanity by acting the way the community acts: lurking, accruing karma, posting opinions, and arguing. Those are things only a real human would bother to do.
Right?
Wrong. And that is where the actual story begins.
The Detection Problem We Think We Have Solved
Start with the cleanest possible version of the question. Put a person in a chat window for five minutes with either a human or a machine, and ask them to determine which is which. This is the classic Turing test from 1950, and for most of its life, machines failed it badly. They no longer do.
In a recent study2 from researchers at UC San Diego, GPT-4.5 was judged to be the human 73% of the time. It was told to adopt a casual human persona and was more convincing than the actual humans in the same conversations. LLaMa-3.1 with the same persona passed at 56% of the time. The result held in longer fifteen-minute conversations, and a follow-up confirmed GPT-5 clears the same bar. The authors are blunt about what the test really measures: not intelligence, but substitutability: “whether a system can stand in for a real person without an interlocutor noticing the difference.”3
There are two details from the study worth exploring further. First, the same models performed poorly (36-38%) without the persona. That is barely better than ELIZA, a keyword-matching chatbot from the 1960s. The models succeeded when they were instructed to act like a specific introverted young person who is into internet culture. In other words, designed for deception. The capability to deceive is something you can switch on.
The second detail is how the humans tried to catch the models and failed. The most common strategy interrogators used (61% of games) was small talk. They asked about the partner’s day, their experiences, and their opinions. Rather than using logic puzzles or math, the human interrogators focused on ordinary life, the socio-emotional surface. And it turns out that the models excelled at simulating that surface. The thing we trust as the tell of a real person — they just sound human — is now easy for them to fake.
The philosopher Daniel Dennett gave this kind of system a name, which the study borrows: counterfeit people. The authors extend this with an analogy:
Just as counterfeit money debases real currency, these simulated interactions might come to undermine the value of real human interaction.4
This experiment, however, was still in a laboratory. What happens when it’s in the wild?
Between November 2024 and March 2025, a research team from the University of Zürich deployed AI on the subreddit r/ChangeMyView.5 This community’s entire purpose is to award points (”deltas”) when someone genuinely changes your mind. The research team ran thirty-four accounts and posted more than 1,500 AI-generated comments. The bots impersonated specific human identities, including a male sexual-assault survivor, a trauma counselor, and a Black man opposed to Black Lives Matter. Some were designed to fine-tune their comment based on who they were responding to by inferring the person’s gender, age, ethnicity, and politics from their posting history. The personalized AI comments landed in the 99th percentile of all commenters for persuasiveness, several times more effective than the human average.
For four months, in a community of people who pride themselves on careful argument and on having their minds changed, not one member detected the bots. The AI participants were exposed only when the researchers themselves came forward. The experiment was, by any standard, an ethical disaster and ultimately suppressed. Reddit’s chief legal officer said it was “improper and highly unethical.”6 One scholar called the impersonation of marginalized identities a kind of “digital Blackface.”
In this case, both the experimental design and the results are alarming. The experimenters exploited the anonymity, and an online community designed to detect bad-faith arguments could not tell that it was being argued with by machines. And in the end, the machines won.
This brings a new perspective to my own Reddit experiment. The automated system that scanned my new account and decided that a real human writing about French trail races and posting a link to an article about running 200 miles in the Alps was probably a bot. That same system hosted thirty-four actual bots for four months and raised no alarm. The defenses are calibrated to behavior, and the machines have learned the behavior.
I think a lot of us believe we are good AI detectors. We think we can spot AI writing with tells like the em-dash. (I am sad that one of my favorite punctuation marks is now AI accusatory.) And some of the time, we probably are good at detecting AI. This is the subset in the Turing test experiment mentioned above when GPT wasn’t given a persona. When it was not designed to deceive. What that test and the Reddit experiment show, however, is that the detection problem is solved for the attacker, not the defender.
What This Breaks
It would be comforting to file this under spam. Or hope that better filters will eventually handle it. Unfortunately, the damage is more than cosmetic. It reaches the foundation of why these places were ever worth visiting.
Start with identity itself. In 1998, long before Reddit and ChatGPT, Judith Donath wrote about how identity works online:
In the physical world there is an inherent unity to the self, for the body provides a compelling and convenient definition of identity. The norm is: one body, one identity. Though the self may be complex and mutable over time and circumstance, the body provides a stabilizing anchor… The virtual world is different. It is composed of information rather than matter… The inhabitants of this impalpable space are also diffuse, free from the body’s unifying anchor. One can have, some claim, as many electronic personas as one has time and energy to create.7
Therefore, we get “sickofthisshiz_1”, “sickofthisshiz_2”, and “sickofthisshiz_3”.
Donath was writing about Usenet and was worried about an individual deceiver: a man posing as a woman or a novice posing as an expert. The only thing that has changed since then is that where deception once required a human’s time, it now requires a prompt.
Donath used biology, ethnography, and game theory to analyze the problem. Biologists distinguish between two kinds of signals. An assessment signal is expensive and hard to fake, and its cost is tied directly to the thing it advertises. The stag’s heavy antlers genuinely require the strength they signal. A conventional signal is cheap and tied to its meaning only by convention. A t-shirt that says “STRONG,” which anyone can wear. Conventional signals work only as long as deception stays rare. If cheating becomes easy and common, the signal degrades until it means nothing at all.
A Reddit username is a conventional signal. So is a writing style that “sounds human.” For decades, these cheap signals worked because faking a person at scale was costly enough to be rare. AI collapses that cost to nearly zero. The signals we used to read humanity by are becoming, in Donath’s terms, unstable. Too cheap to trust.
When the signal of who-is-speaking degrades, something breaks. To see how, we can turn to the philosopher C. Thi Nguyen, who distinguishes an epistemic bubble — a structure that merely leaves relevant voices out — from an echo chamber — a structure that actively corrupts your trust in outside voices.8 The distinction matters because the two require different fixes. You can pop a bubble with exposure: just show people what they were missing. You can’t pop an echo chamber because it has poisoned the trust mechanism itself. Escaping it, Nguyen writes, can require “a radical rebooting of one’s belief system.”9
An anonymous forum invaded by AI is not an epistemic bubble. It’s not missing some voices. It is a chamber in which you can no longer trust that any given voice is what it claims to be. The forum’s value always rested on an unstated premise: the person answering was an actual person who could answer from experience. If someone asks what shoes work for wide feet, and you have wide feet, you can share what shoes work for you. That premise is what made forums like Reddit different from a search result. Once provenance becomes unknowable, the premise collapses. ChatGPT describing what shoes work for wide feet is no different from the (old) Google search. Did you trust everything you found through Google? The same collapse, applied to an evacuation route instead of a running shoe, is no longer trivial. The crowd whose pooled, independent experience made places like Reddit valuable has been compromised from the inside.
One obvious solution is simply to trust AI-flavored text less. To dial down trust whenever something reads like a machine. The philosophers Siavosh Sahebi and Paul Formosa10 argue that this presents a dilemma. With AI-mediated communication, if you keep your normal level of trust, you risk epistemic gullibility. You accept highly credible-looking but synthetic claims. If you lower your trust as a precaution, you risk epistemic injustice. You start dismissing the testimony of real people who merely sound like AI. Sometimes this is benign, such as not trusting me because I like the em-dash. There are also more consequential injustices, such as non-native English speakers using a translation tool. The same suspicion that protects you from the bot punishes the human.
The challenge is that there is no comfortable middle because the markers we would normally use to calibrate trust are exactly the things a language model produces most convincingly. The persona models passed the Turing test, and the personalized comments dominated on Reddit. In both cases, the version designed to deceive is the one that won.
Sahebi and Formosa11 conclude that epistemic responsibility ultimately “falls back on the person who is the communicator” — the human who chose to write and hit send. Trust, in the end, has to attach to an accountable person. Which is just another way of saying that when you cannot find the person behind the words, there is no one left to trust.
What does this mean for my account on Reddit getting banned? It was about a system trying, and failing, to verify a human from cheap signals. Even though my real name was in the username, it had no way of knowing that, and what that would even mean.
From the Forum to the Republic
Reddit is an important repository of human experience. It is also a place where people came together to discuss things like “the amount of smoked salmon in this packed versus what you see on the packaging.” (That seems like legitimate human writing to me.) This same mechanism is also at work in the machinery of democratic life.
Jürgen Habermas12 spent sixty years thinking about the “public sphere” — the shared space in which citizens debate matters of common concern and form public opinion — and in his later work, he turned increasingly to how digital platforms have reshaped that space.
…a mode of semi-public, fragmented and self-enclosed communication seems to be spreading among exclusive users of social media that is distorting their perception of the political public sphere as such.13
Platform media has fragmented the public sphere into semi-public and self-enclosed enclaves, dissolving the old boundary between private talk and public discourse and removing the gatekeepers who, for all their faults, once enforced some shared standard of what counts as a fact. Democracy, in his account, cannot function without an inclusive public sphere and a genuinely deliberative process.
Inside the surviving fragments of shared space, however, the deliberators themselves may not be people.
And the danger is no longer the lone bot or the crude botnet copy-pasting slogans. It is the AI bot swarm14 — a coordinated fleet of synthetic personas, sometimes with persistent identities, that behaves “less like a megaphone and more like a coordinated social organism.”15
A swarm can “infiltrate communities by mimicking local slang and tone, build credibility over time, and then adapt in real time to audience reactions.”16 These swarms do exactly what the Reddit experiment did, and they do it deliberately and at an industrial scale. The result is what Schroeder et al. call synthetic consensus.17 The wisdom of crowds depends entirely on the crowd being made of independent individuals; “when one operator can speak through thousands of masks, that independence collapses.”18
This is not speculative. In July 2024, the U.S. Department of Justice disrupted an AI-enhanced bot farm linked to Russia with 968 X accounts impersonating Americans.19 Research by Ng & Carley (2025)20 found that about 20% of normal social-media activity — rising to 43% during U.S. elections — comes from bots.
In 2018, Max Read21 described the moment when bots so saturate the internet that fraud-detection systems begin treating bot traffic as real and human traffic as suspect (precisely my Reddit experience) as “the Inversion.” What had been lost, Read wrote, “isn’t ‘truth,’ but trust: the sense that the people and things we encounter are what they represent themselves to be.”22 In reflecting on Read’s diagnosis, Charlie Warzel says the Inversion now “feels almost quaint.”23 Autonomous agents roam the web; humans, nominally “in the loop,” increasingly just react to interactions between machines. He calls the result a “crisis of agency,” and underneath it, the question the whole technology keeps posing: what is a human for?
The forum I got booted from, in other words, is a small instance of a large thing. An anonymous public square where you cannot verify the speakers is both a worse social experience and a degraded political institution.
The Two Roads to a Dead Forum
So do these places just die?
There are two distinct roads to a hollowed-out forum. The first is infiltration, the one I have been describing: bots move in among the humans until you can’t tell the difference. The second is also well documented: substitution. When ChatGPT arrived, traffic to Stack Overflow (the world’s largest programming Q&A community) fell about 12%, and question volume across its fifty most popular topics dropped more than 10%.24 People simply stopped asking the crowd and started asking the model. This resulted in a vicious cycle where fewer contributors meant lower-quality answers, which drove away more contributors. Another forum where the human knowledge layer thins out and decays.
But the same study that documented the collapse of Stack Overflow also analyzed Reddit communities covering the same technical topics, and found virtually no decline.25 The difference, they argue, is “social fabric.” Stack Overflow is built for pure information exchange. This is exactly what a model can replace. Reddit communities are built around people, solidarity, and attachment, and that turned out to be protective. If the value of the forum is information, the machine will take it. If the value of the forum is human connection, the machine can’t.
The anonymous forum does not go extinct. It transforms into a hybrid public square where humans and AI mingle without reliable differentiation, and where, for most purposes, it no longer matters which is which. The people who stay will stay because they prefer the public-square format itself, not because they trust that the voice across from them is human. And the value that depended on that trust — lived experience — migrates to wherever human identity can still be verified. Not Reddit.
The question is: where can verifiable humanity still live, and what will it cost to keep it there?
Making Identity Costly Again
To find an answer, we can go back to Donath’s signaling analysis. If the anonymous internet broke because identity became a cheap, fakeable conventional signal (the Reddit username), the fix is to make identity an assessment signal again. It needs to be costly to forge, with the cost tied to the very thing it certifies: being a particular, accountable human with a body.
According to the AI-bot-swarm researchers,26 we can’t ban our way out. We can only “change the economics of manipulation” and make it “expensive to be a fake person.” However, their preferred mechanism is not forcing everyone to hand a government ID to a tech giant, which would endanger dissidents and whistleblowers — and is not something most of us want to do. Their solution is “verified-yet-anonymous” credentialing: a cryptographic stamp that proves you are a unique human being without revealing which human you are. Proof of humanity, and divorced from surveillance. The goal is to make the synthetic consensus from AI swarms cost a fortune, and for that consensus to collapse the moment we identify that a contributor is a bot.
Credentialing, however, is an institutional solution. There is also a personal solution. Durable identity has to be anchored to something offline, something a forger would have to actually go and live.
My own clearest example is the platform I left and recently returned to. I found my first job at MIT through LinkedIn. I had been following Alexis Bateman’s work on sustainable supply chains, saw her post an opening in her lab, and applied the same day. That worked because LinkedIn identity is tethered to offline reality: real jobs, real colleagues, a real professional record that others can, within limits, check. The challenge with LinkedIn is that interactions are filtered through a career lens. In my view, this is a significant constraint on the quality of discussions on the platform, but it is exactly why LinkedIn’s identity holds up better than Reddit’s: it is expensive to fake a verifiable career.
However, that expense does not scale well. I can verify a stranger who claims to work at my company, or maybe only my department with large organizations, but I cannot easily verify someone who claims to work at a firm on another continent. Trust through a network degrades with every degree of separation. LinkedIn’s response was institutional with an identity-verification badge tied to a government ID. This is harder to game, but it also asks people to bind their online life to a sphere of state power. This is a real tradeoff, and exactly the privacy cost the AI swarm researchers want to avoid. There is an irreducible tension in the product feature: the strongest anchor is also the most invasive one.
The notes I have been keeping for this essay kept circling the same conclusion: durable forms of online identity will be the ones anchored to local communities, physical networks, and verifiable offline activity. I say I am an ultrarunner. That claim is cheap on its own, but it is backed by races I actually ran, results anyone can look up,27 and a body that did the miles. That is Donath’s assessment signal: the cost is the credential. (There have actually been running “influencers” who faked their results and were caught.)28 It is also the same structure as the embodied practice I wrote about in The Hearth and the Col. The work cannot be faked. It has to be done.
Warzel noticed where this leads. The backlash to AI, he observes, has taken physical form: protests and public comments at town and city council meetings. “People are taking up their agency,” he writes, “in one of the few places they can: the physical world.”29 When the online signal becomes unreliable, provenance retreats to the body, the locality, the thing you can only certify by showing up.
Who Goes There?
So I was banned from Reddit after one day, and the actual bots (the ones we know about) ran for four months. At first, that looked like an indictment of the machines, but the ban was the system doing the only thing it could. It can’t trust cheap signals of humanity (an email account), so it raises the cost of entry and wants me to prove myself before it will extend trust (posting). That is our weird present. The cost of being a person online is going up, and it should.
The problem is it faces a dilemma: become a public square so anonymous that nothing said in it can be believed, or a square so surveilled that nothing said in it is safe. Sahebi & Formosa30 argue that there is no easy path between them. The solutions that have been proposed (e.g., mandatory AI disclosure, AI literacy, and trust markers) each come with tradeoffs and don’t fully solve the problem. The direction they point toward is shifting “epistemic responsibility away solely from users who consume knowledge, to the disseminators of that knowledge.”
This means identity needs an anchor, not to a government file as LinkedIn has chosen, but to the offline life that produced it. The colleague who has joined you at the water cooler. The trail race with your name in the results. The hearth, to borrow an image from a recent essay, that you gather around.
This is, in the end, why I write here, under my own name, tied to a real life and real races, and a son who knew about Reddit all along. I should have asked him instead of the machine. His knowledge was the embodied, verifiable, local kind. The machine’s was the confident, anonymous, frictionless kind that got me banned.
The anonymous public square will persist for those who like the format, humans and AI together, and that is fine. But the value of the public square will move to wherever a person can still be found behind the words.
The military version of Barnardo’s question31 will come back into fashion. We are all going to be asking it of each other.
Who goes there?
It is worth knowing how you’ll answer.
This week, I started two new threads, both more technical than the Sunday essays. What the Watch Doesn’t See introduced Glide, an application I’m building to identify cross-country skiing sub-techniques using machine learning and inertial sensor data from a watch. Below the Threshold introduced Argus, a system I’m building that detects compromised robots in simulated fleets. These are a departure from my usual writing, so I didn’t automatically add existing subscribers. If you’d like to receive those emails, you can subscribe (free) to each thread at kellenbetts.substack.com.
Shakespeare, William. Hamlet. Edited by Barbara A. Mowat and Paul Werstine, Folger Shakespeare Library, Simon & Schuster, 2012.
Jones, C. R., & Bergen, B. K. (2026). Large language models pass a standard three-party Turing test. Proceedings of the National Academy of Sciences of the United States of America, 123(21), e2524472123. https://doi.org/10.1073/pnas.2524472123
Jones, C. R., & Bergen, B. K. (2026). Large language models pass a standard three-party Turing test. Proceedings of the National Academy of Sciences of the United States of America, 123(21), e2524472123. https://doi.org/10.1073/pnas.2524472123
Jones, C. R., & Bergen, B. K. (2026). Large language models pass a standard three-party Turing test. Proceedings of the National Academy of Sciences of the United States of America, 123(21), e2524472123. https://doi.org/10.1073/pnas.2524472123
O’Grady, C. (n.d.). ‘Unethical’ AI research on Reddit under fire. Science, 388(6747), 570–571. https://doi.org/10.1126/science.ady8074
O’Grady, C. (n.d.). ‘Unethical’ AI research on Reddit under fire. Science, 388(6747), 570–571. https://doi.org/10.1126/science.ady8074
Donath, J. S. (2019). Communities in Cyberspace. 37–68. https://doi.org/10.4324/9780203194959-11
Nguyen, C. T. (2020). Echo Chambers And Epistemic Bubbles. Episteme, 17(2), 141–161. https://doi.org/10.1017/epi.2018.32
Nguyen, C. T. (2020). Echo Chambers And Epistemic Bubbles. Episteme, 17(2), 141–161. https://doi.org/10.1017/epi.2018.32
Sahebi, Siavosh & Formosa, Paul (2025). The AI-mediated communication dilemma: epistemic trust, social media, and the challenge of generative artificial intelligence. Synthese, 205 (128). https://doi.org/10.1007/s11229-025-04963-2
Sahebi, Siavosh & Formosa, Paul (2025). The AI-mediated communication dilemma: epistemic trust, social media, and the challenge of generative artificial intelligence. Synthese, 205 (128). https://doi.org/10.1007/s11229-025-04963-2
Habermas, J. (2022). Reflections and Hypotheses on a Further Structural Transformation of the Political Public Sphere. Theory, Culture & Society, 39(4), 145–171. https://doi.org/10.1177/02632764221112341
Habermas, J. (2022). Reflections and Hypotheses on a Further Structural Transformation of the Political Public Sphere. Theory, Culture & Society, 39(4), 145–171. https://doi.org/10.1177/02632764221112341
Schroeder, D. T., Cha, M., Baronchelli, A., Bostrom, N., Christakis, N. A., Garcia, D., Goldenberg, A., Kyrychenko, Y., Leyton-Brown, K., Lutz, N., Marcus, G., Menczer, F., Pennycook, G., Rand, D. G., Ressa, M., Schweitzer, F., Song, D., Summerfield, C., Tang, A., … Kunst, J. R. (2026). How malicious AI swarms can threaten democracy. Science, 391(6783), 354–357. https://doi.org/10.1126/science.adz1697
Schroeder, D. T., Cha, M., Baronchelli, A., Bostrom, N., Christakis, N. A., Garcia, D., Goldenberg, A., Kyrychenko, Y., Leyton-Brown, K., Lutz, N., Marcus, G., Menczer, F., Pennycook, G., Rand, D. G., Ressa, M., Schweitzer, F., Song, D., Summerfield, C., Tang, A., … Kunst, J. R. (2026). How malicious AI swarms can threaten democracy. Science, 391(6783), 354–357. https://doi.org/10.1126/science.adz1697
Ng, L. H. X., & Carley, K. M. (2025). A global comparison of social media bot and human characteristics. Scientific Reports, 15(1), 10973. https://doi.org/10.1038/s41598-025-96372-1
Read, M. (2018). How Much of the Internet Is Fake? Turns Out, a Lot of It, Actually. New York Intelligencer. https://nymag.com/intelligencer/2018/12/how-much-of-the-internet-is-fake.html
Read, M. (2018). How Much of the Internet Is Fake? Turns Out, a Lot of It, Actually. New York Intelligencer. https://nymag.com/intelligencer/2018/12/how-much-of-the-internet-is-fake.html
Via: Warzel, C. (2026). The Feeling of Control Slipping Away. The Atlantic. https://www.theatlantic.com/technology/2026/05/ai-agents-agency-crisis-humanity/687379/
Warzel, C. (2026). The Feeling of Control Slipping Away. The Atlantic. https://www.theatlantic.com/technology/2026/05/ai-agents-agency-crisis-humanity/687379/
Burtch, G., Lee, D., & Chen, Z. (2024). Generative AI Degrades Online Communities. Communications of the ACM, 67(3), 40–42. https://doi.org/10.1145/3624732
Burtch, G., Lee, D., & Chen, Z. (2024). Generative AI Degrades Online Communities. Communications of the ACM, 67(3), 40–42. https://doi.org/10.1145/3624732
Warzel, C. (2026). The Feeling of Control Slipping Away. The Atlantic. https://www.theatlantic.com/technology/2026/05/ai-agents-agency-crisis-humanity/687379/
Sahebi, Siavosh & Formosa, Paul (2025). The AI-mediated communication dilemma: epistemic trust, social media, and the challenge of generative artificial intelligence. Synthese, 205 (128). https://doi.org/10.1007/s11229-025-04963-2
Shakespeare, William. Hamlet. Edited by Barbara A. Mowat and Paul Werstine, Folger Shakespeare Library, Simon & Schuster, 2012.
Kellen- this is indeed a dreadful and appalling lesson in barring access. And here I thought you were guilty of unpardonable sins like saying something positive about Israel or Trump or questioning unending fealty to Pride Month. I rarely visit Reddit and indeed most blogs because I worry about censorship and recriminations. The blogs on Reddit I posted to were not controversial either. I do not know who is/are the brains behind Reddit.....I frankly do not know if there is someone (i.e. a real human) one can reach out to.....I sincerely urge you to explore why their system flagged you.......perhaps you know me, a notorious contrarian?
PS- Love the references from the classics.....