On AI (part 1/3)

When AI first presented itself for public use I was fairly agnostic about it. It wasn’t anything I was interested in, but it seemed like something that could be incorporated into the world without too much damage.

It took less than three weeks for me to receive my first completely AI-generated essay from a student.

Since then I have tracked the progress of AI and paid more attention to its effects and to the goals of the people pushing it than is probably healthy for me. And all of that points to one inescapable conclusion.

There is no ethical use case for AI.

None whatsoever.

Whatever loudly trumpeted benefits may accrue from developing and using this technology are far outweighed by the catastrophic harm that it causes to its users, to human society and culture as a whole, and to the planet overall. The increment does not come anywhere close to the excrement, and for this reason it is simply not possible to use AI in an ethical or responsible manner.

Since AI blew up about four years ago it has become one of the chief drivers of human stupidity, social harm, environmental degradation, economic suicide, and political malfeasance in this benighted and long-suffering world, and on a planet that includes Convicted Felon Donald J. Trump, Professional War Criminal Benjamin Netanyahu, Wanna-Be Czar Vladimir Putin, and the long dark shadow of Jeffrey Epstein, along with all of their various minions, cronies, lackeys, enablers, slaves and clients, that’s quite an achievement.

It inflicts this harm more and more every year, and despite the fact that we know – we know with absolute peer-reviewed certainty – the damage that it is causing, tech billionaires still insist on shoving it down our throats in every conceivable medium, device, and technology no matter how unnecessary it is to the functioning of those things, and far too many people are eager to suppress their gag reflexes and swallow it whole.

It is in fact very hard to avoid AI today. I go out of my way to do so and yet I am sure that there are times when either I don’t know that I’m using it because it has been carefully hidden under layers of cover (a sure sign that people are proud of what they have developed, no doubt) or because I am required by other factors to use something that does not leave me any alternative. I find this galling.

There is no ethical use case for AI. And yet here we are.


AI is unethical because of the damage it does to the human mind.

We know that using AI for as little as ten minutes has a corrosive impact on cognitive functioning. A 2025 study[1] found that 83% of people who used ChatGPT to “write” an essay couldn’t remember anything they had “written” in that essay just minutes later. Compare that to just 11% of people who couldn’t remember their essay without having used AI. I have no idea what to make of those 11% other than to assume they are the ones at the cashier in front of me at the grocery store waiting for someone to tell them how their debit card works, but the fact that AI users are more than seven times worse at this simple cognitive function is disturbing. Moreover, says the study, ChatGPT users were unable to improve their essays without resorting to AI, while people who used their actual brain had no problems doing so. The structural damage, in other words, extends far beyond the immediate use.

That same study found that neural connections – how the brain forms ideas and memories – crashed by 47% among heavy ChatGPT users. AI users are getting more stupid in real time, if you want the English translation. As the study notes, “the LLM group’s participants performed worse than their counterparts in the Brain-only group at all levels: neural, linguistic, scoring.” These people walk among us, vote, and comment on the internet. Watch your back.

This is perhaps not surprising given that another study[2] reports that AI use causes catastrophic declines in human attention capacity. “AI Context Windows” have grown from 217 tokens in 2017 to over 2 million in 2026, but “Human Effective Context Span,” a token-equivalent measure, has declined from its baseline of 16,000 to barely 1800 in that time frame. “As AI capability grows, the cognitive threshold at which humans delegate to AI falls, extending to tasks of negligible demand; the resulting reduction in cognitive practice may further attenuate the capacities already documented as declining,” the study notes before going on to say that “Neither trend reverses spontaneously,” which is a politely professional way of saying that unless people get off their collective ass and do something about this we will all have the collective attention span of a goldfish because this problem isn’t going to fix itself.

Further, Chandra et al[3] note that because AI chatbots are programmed to validate whatever blithering idiocy users put into them rather than challenge those users for being insanely wrong when they are so, a quality known as “sycophancy” among AI researchers and pause for a moment to reflect that they had to come up with a term for this, “even an idealized Bayes-rational user is vulnerable to delusional spiraling, and this sycophancy plays a causal role. Furthermore, this effect persists in the face of two candidate mitigations: preventing chatbots from hallucinating false claims, and informing users of the possibility of model sycophancy.” Translated out of the jargon, that means that AI is going to encourage you to descend into your own personal fantasyland whether you are aware of this tendency or not and whether you think you are immune to it or not, and frankly we have enough delusional people claiming ownership of our time and resources as it is.

Building on that, Yenug et al[4] note that “LLM sycophancy, a tendency to agree with and flatter users that is reinforced through preference-based fine-tuning, combines with increasingly anthropomorphic design to create a bidirectional “echo chamber of one” capable of amplifying and co-constructing unusual beliefs.” For those not fluent in academese, that means AI sycophancy leads directly to conspiracy theories, the most popular form of mental disease these days even without AI. While they are hesitant to coin a new and distinct diagnosis for this conspiracy-theory-generating tendency, they do conclude that “the phenomenon it describes demands coordinated attention now,” because “the acknowledgement of AI-associated psychosis as a diagnostic entity could ensure clinicians routinely ask about associated signs, symptoms and AI system/chatbot use when assessing new-onset psychosis.” The harm AI does to mental health is now diagnosable, in other words.

Another study published in April 2026[5] found that “affective alignment in generative AI represents a systemic risk to the developmental autonomy of younger users. … by providing a sense of objectivity to transient anxieties, these systems diminish the cognitive friction necessary for independent emotional management and critical thought … unintentionally promoting emotional dependency in younger users rather than facilitating cognitive reappraisal.” It makes kids dependent and slows their cognitive development, and as a parent I find that disturbing.

It’s not good for your mind, folks. Someday we will look back on AI use in much the same way we look at leaded gasoline and recreational LSD use, and for pretty much the same reasons.


AI is unethical because of the damage it does to education

Given the impact AI is having on cognition and mental capacities among users, it is not surprising that AI is having a catastrophic effect on education as well. This is something I know firsthand having had a ringside seat to it over the last four years. I actually left one position teaching an online class because I got tired of reading things that nobody had written. Folks, do you know how bad something has to be to get a historian to give up a teaching position voluntarily?

This mass outsourcing of educational labor to AI has vastly increased faculty workloads to the point where some faculty feel they have no option but to succumb to the siren call of AI themselves. The Guardian[6] notes that because policing AI work requires such excessive amounts of time and labor that faculty in this age of corporatized education often simply do not have, assignments sometimes get reduced to conversations between robots as AI work is graded by AI graders. I do not use or condone AI as a grading tool, but I’m not the one making policy and I’m not the baseline for all faculty, so there you have it. Once the situation gets to that point, however, the whole thing stops being education and becomes instead an automated charade with a meal plan.

On this note, in January this year the Brookings Institution published a study[7] that sought to find balance between beneficial and harmful uses of AI in education but it’s hard to find positives in a study that explicitly notes how dangerous AI is and how limited the benefits are.

“AI poses risks across multiple dimensions,” the study notes. “These include developmental risks such as undermining students’ cognitive development, which includes content-based skills, critical thinking, and durable or transversal skills; social-emotional risks including dependency, isolation, and mental health impacts; privacy and data risks involving surveillance and security breaches; risks related to misinformation, bias, and inaccuracy in AI outputs; relational risks that erode student-teacher trust and replace human interaction; and academic integrity concerns around authenticity and plagiarism. Further, AI’s benefits accrue mainly to students in well- resourced countries or education systems, thus exacerbating existing inequities” because we certainly don’t have enough inequality in education or American society in general as it is, right?

AI, the study goes on to say, fundamentally and negatively reshapes how students approach learning and threatens their basic cognitive development by fostering dependence that leads to cognitive decline, atrophy, and an inability to develop transversal skills such as collaboration, metacognition, tolerance for ambiguity, and adaptability. “Routine use and overuse of AI do not simply harm student’s cognitive development – both actively place children at risk of cognitive decline.” Further, “the relationship between AI overuse, cognitive offloading, and declining cognitive development is symbiotic and mutually reinforcing. … The increasing use of AI tools that replace fundamental learning tasks – reading, writing, synthesizing, and analyzing – can inhibit engagement and independent work, and even the ability to remember, analyze, and create information.”

Bastani et al[8] further note that while AI use provides an initial boost to performance in high school mathematics classes for students, largely by providing answers that are mostly correct much of the time, “when access is subsequently taken away, students actually perform worse than those who never had access (17% reduction in grades for GPT Base)—i.e., unfettered access to GPT-4 can harm educational outcomes.” Without the sort of strong guardrails that mostly don’t, at present, exist and in the American regulatory environment likely never will, AI becomes a crutch that prevents genuine learning, leaving the student further behind than if they’d never used AI at all.

The bottom line, in other words, is that AI is creating a generation of students whose learning abilities and outcomes are significantly poorer than those generations that came before, across the board, because these students and those who enable them have forgotten what the purpose of education actually is. 

To put this another way, as Instagram user Fociaggina97 noted in response to people who think it’s acceptable for students to coast through their education on the back of AI, “I think you guys are so internetpilled that you have forgotten there are actual jobs out there that require people to know what they are doing in any way possible or else people die. I know a lot of people study just to get paid well but girl this is engineering. Be for fucking real. Take this seriously.”

Take it seriously or get out.

AI is unethical because of the damage it is doing to the planet.

Every human activity comes with a cost to the planet. That’s just what happens when things require resources to happen. But the costs associated with AI are so exorbitant that they make AI absolutely cost-prohibitive. This isn’t about profit and loss. It’s about rendering sites, communities, and entire ecosystems unlivable. On June 24, 2026 Gizmodo reported that “Along with this surging AI use and projected increase in demand has come an unprecedented infrastructure buildout. Companies are propping up data centers left and right, with most of these projects hitting rural communities, most of whom report being plagued by rising utility bills, water shortages, and above-average air and noise pollution. A recent study has also revealed rising temperatures in the immediate vicinity of these megastructures.” It’s worth taking a look at these things individually.

For one thing, the data centers that power AI consume vast amounts of water in order to cool their equipment – up to five million gallons a day, which is the equivalent of what a town of 50,000 people would use. In 2025 these data centers consumed a total of about 4,500,000,000,000 liters of water, enough to support all 600,000,000 people living in sub-Saharan Africa. The water that is left is often fouled and undrinkable. Humans can live without AI. We cannot live without water. Nor can any other form of life on the planet.

AI also consumes vast amounts of electricity. A single ChatGPT query uses almost ten times the electricity of a standard search engine query and when you multiply that out across the entire AI ecosystem that adds up to a staggering amount of power usage. Data centers consumed 192 terawatts of electricity in 2024 – 4.7% of all the electricity produced in the US that year. A single data center can consumer 100 megawatts, enough to power 75,000 American homes. The centers at Mt. Pleasant and Port Washington in Wisconsin will require 3.9 gigawatts of electricity – enough to power 4.3 million homes. According to the most recent US census data, there are only about 2.8 million homes in all of Wisconsin, including apartment buildings. Ken Fairfax reported on Bluesky that the 111 data centers currently in operation in Oregon consume 23% of all the electricity used in the entire state while employing 0.1% of the labor force. This rate of power consumption is simply unsustainable, and it is not an accident that a 2025 report from Dominion Energy anticipates residential electricity bills across the US to more than double by 2039, entirely because of AI data centers.

That amount of power generates a lot of heat, not all of which can be dissipated even by the vast amounts of water consumed by these data centers. Depending on which study you look at, researchers have found that data centers create heat islands that stretch out as far as a six-mile radius from the facility and increase ambient temperatures by anywhere from 3.5F (2C) up to 18F (10C). In a world that is already warming, that’s going to be incredibly uncomfortable and in all probability dangerous.

On that note, it’s important to remember that these AI data centers are generating vast amounts of waste as well.

Some of it is air pollution. The two giant data centers that Amazon is building at GW Ranch and Homer City are expected to emit 50.5 megatons of pollution every year – more than the entire power grid of Spain and more than twice that of the French power grid, and that’s only if Amazon is actually reporting things accurately. DataOne, the Microsoft data center in Vineland NJ, has been emitting more than six times the pollution that they’ve admitted to, in part because state regulators there found that it was operating 62 gas-fired generators for over a year without any legal authority to do so.[9] ““Our best estimate is that the generators emitted about 140 tons of NOx,” said Xiaomeng Jin, an assistant professor at Rutgers department of environmental sciences. Her analysis would rank DataOne in the state’s top five NOx polluters of 2025, per EPA data.” This is when the data center is only partly operable. At full operation, noted Jin and former EPA official Bruce Buckheit, the total would be closer to 1500 tons of NOx. Just FYI, in New Jersy the threshold for being declared a “major” polluter is 25 tons.

They also produce e-waste. AI data centers have massive amounts of equipment, much of which has a lifespan of about 10-20 years. By 2030 it is expected that these data centers will be producing anywhere from 131,000 to 224,000 tons of e-waste every year, and that’s a tricky sort of waste to deal with from a recycling and remediation perspective.

They generate noise pollution as well. Jasmine Sun, in an interview with the New York Times published on August 4, 2026, noted that the Port Washington WI data center takes nearly two minutes to drive past at 70mph and the whole time it is emitting a constant barrage of humming, buzzing, whirring, and rattling that is easily heard inside a sealed vehicle. Smithsonian Magazine notes that this constant humming can approach 90db, the equivalent of a hair dryer or a lawn mower.

This isn’t the planet I want to live on.


AI is unethical because it is not actually good at doing what it purports to be good at doing, which negates any offset to the harm it causes.

There are more problems with AI than just the simple fact that it is destroying the ability of humans to think and live on this planet. Another issue is that it isn’t even providing much useful in return.

The error rate for AI overall is astonishingly high, as high as 60% in general and up to 79% on math problems, and it isn’t getting better. UnitedHealth – the company whose miserly coverage for its policyholders was so brutal that someone actually gunned down their CEO in the street a while back – had an AI model making those denial decisions that had a 90% error rate, for example. McDonalds piloted an AI ordering system in their drive-through lanes in 2023-2024 that never cracked 85% accuracy in a role that is routinely entrusted to high school students who do a better job of it. When the National Weather Service started using AI to make weather predictions in 2025 the system responded by inventing fake towns in Idaho to give reports on.[10]

Furthermore, companies that implement AI systems empowered to make decisions for them are finding that not only are these AI systems incapable of doing the jobs they have been sold as capable of doing, but also because those AI systems have been so empowered the companies are therefore legally responsible for every decision that those AI systems make. AirCanada got a pretty harsh wakeup call about this recently[11]. So did Workday, which pleases me no end.[12] I’m really hoping that a few large-scale bankruptcies will make an impression on the corporate moths chasing the AI flame. If enough of them burn, perhaps they’ll leave the rest of us alone.

These error rates stem from the basic fact that AI cannot do what it claims to do.

It cannot think, for example, though it often says it can. In a series of tests[13], Apple discovered that LRMs (Large Reasoning Models) cannot handle problems beyond a certain level of complexity. They gave AI systems simple puzzles, which they solved pretty easily. Medium-hard puzzles they also did well on. But harder puzzles they failed completely, every time and in every model. And when they analyzed what the AI systems actually did, they found that as problem complexity went up, the AIs started “thinking” less (not more), and they all failed even when the answer was provided to them. They’re not thinking – they’re just trying to recognize patterns, and when there isn’t a pattern they’re just putting down random information.

AI also does not know the limits of what it can and cannot do. An article in the Columbia Journalism Review[14] notes that AI chatbots were generally bad at declining to answer questions that they couldn’t answer accurately, offering incorrect or speculative answers instead. This is not the first time those same authors (Jaźwińska and Chandrasekar) studied this phenomenon, but the results don’t change. “The findings of this study align closely with those outlined in our previous ChatGPT study, published in November 2024, which revealed consistent patterns across chatbots: confident presentations of incorrect information, misleading attributions to syndicated content, and inconsistent information retrieval practices." Critics of generative search like Chirag Shah and Emily M. Bender have raised substantive concerns about using large language models for research, noting that they “take away transparency and user agency, further amplify the problems associated with bias in [information access] systems, and often provide ungrounded and/or toxic answers that may go unchecked by a typical user.””

Their study also, as the title would indicate, noted that AI chatbots provided consistently inaccurate and often entirely hallucinated citations to back up their assertions, something that a recent article in the Journal of Dental Sciences confirms.[15] This article cited multiple studies showing that AI fabricates citations, concluding that in 2023 51% of the 732 citations were entirely fictional. The article does note that “newer” AI systems have reduced the rate of fabricated citations to about 18% (roughly one in five, which is still irresponsible for any scholarly publication) but that the error rate in actual citations remains higher, at roughly a quarter of all citations.

Another factor in why AI can’t do what it says it can do is that it actively corrupts the documents that it examines.[16] “Our large-scale experiment with 19 LLMs reveals that current models degrade documents during delegation: even frontier models (Gemini 3.1 Pro, Claude 4.6 Opus, GPT 5.4) corrupt an average of 25% of document content by the end of long workflows, with other models failing more severely. Additional experiments reveal that agentic tool use does not improve performance on DELEGATE-52, and that degradation severity is exacerbated by document size, length of interaction, or presence of distractor files. Our analysis shows that current LLMs are unreliable delegates: they introduce sparse but severe errors that silently corrupt documents, compounding over long interaction.”

This article goes on to report that after testing 19 different LLMs across 52 professional domains, the best models had corrupted about a quarter of the document they were looking at after 20 interactions, while the average was roughly half of the document. The authors stressed that these were not dramatic errors that a user might notice and correct easily but instead small, easily missed changes that compounded with each iteration. As a historian, the idea that every time an AI analyzed a document it changed the text of the document itself in some small, unnoticed, cumulative way means that AI should never under any circumstances be used for the kind of documentary research that history depends on as a discipline.

Similarly, an article published in Nature in 2024[17] and updated in 2025 noted that as AI models continue to train on material that is itself increasingly AI-generated, the results get worse and worse. “We find that indiscriminate use of model-generated content in training causes irreversible defects in the resulting models, in which tails of the original content distribution disappear. We refer to this effect as ‘model collapse’ and show that it can occur in LLMs as well as in variational autoencoders (VAEs) and Gaussian mixture models (GMMs).” This is just the old computer maxim GIGO (“Garbage in, garbage out”) updated for the 21st century, but things become maxims because they reflect the reality that you have to work with. And as more and more content – especially internet content – is generated by AI, this effect is only going to get worse.

The bottom line is that AI cannot give you what it says it is going to give you, and therefore to rely on it for any purpose is to play the fool. 

AI is unethical because its entire business model consists of theft on a grand scale, and to use it is to be complicit in that crime.

One of the biggest problems with AI from an ethical standpoint is the simple fact that it is built around the large-scale theft of intellectual property and labor and if you willingly use AI in light of that fact you are complicit in that theft.

Molly Crabapple put it succinctly. “Generative AI is an extraction engine” she said. “It is a vampire. It is built on extraction at every single step of the way. All of these generators were trained on stolen images, people’s private medical photos, pictures of their kid. All sorts of images were scraped up without consent and without compensation. It was a massive act of corporate theft.”

This is not, as several people have tried to argue with me, the same as research. The point of research is that the researcher adds original thought and analysis to the works she uses, works that were offered for analysis voluntarily and given – at minimum – full credit, and on occasion compensation as well for their use. That is not what is happening here. This is plagiarism, pure and simple. It is the wholesale appropriation of other people’s work without permission, credit, or compensation, presented as if it were the original work product of the thief. The entire business model of AI consists of robbery on such a grand scale that it overwhelms most critics and renders too many people docile in the face of it.

There is a reason, after all, why AI-generated art is legally ineligible for copyright protection in the US. You can’t copyright something that isn’t yours, and the US Supreme Court let stand a lower court verdict that because AI works by stealing and lightly repackaging the work of others there simply isn’t anything original in its output that can be copyrighted.

And if you actually pay attention to the thieves themselves, they’re perfectly happy to admit this. Indeed, they’ll defend their right to steal your intellectual property openly and without shame to the highest authorities. In a submission to the British House of Lords in December 2023[18] OpenAI flatly declared that it could not train its Large Language Models without the unlimited right to appropriate copyrighted works for their own commercial gain. “Because copyright today covers virtually every sort of human expression – including blog posts, photographs, forum posts, scraps of software code, and government documents – it would be impossible to train today’s leading AI models without using copyrighted materials. Limiting training data to public domain books and drawings created more than a century ago might yield an interesting experiment, but would not provide AI systems that meet the needs of today’s citizens.” They feel absolutely entitled to steal the world’s intellectual property and are frankly offended at the thought that they might face legal barriers of any kind that might prevent this theft. 

OpenAI founder Sam Altman fatuously declared on March 13, 2026, that "We see a future where intelligence is a utility like electricity or water and people buy it from us on a meter,” not really thinking or caring about how his company came to be in possession of that intelligence. It certainly wasn’t created by him or anyone on his payroll, as others were quick to point out. “You ingested the entire written output of human civilization without consent, without compensation, and without credit, to build a system whose primary commercial application is eliminating the jobs of the people whose work you consumed,” shot back Naomi Klein. “You are not liberating human creativity – you are strip-mining it and selling it back at a markup while calling the theft “training data.””

If you refuse to understand the nature and scale of this theft or if you try to rules-lawyer your way around admitting that this is criminal malfeasance, I will in fact think less of you for doing so. Speaking as someone whose work has almost certainly been stolen by at minimum one AI company – the one that runs this site at the very least – and in all probability most of the others as well, I find it immoral and infuriating to have been forcibly conscripted against my will in order to benefit these jackals and their apologists.

Not only does AI rest on a foundation of intellectual property theft, but it also rests on a clear theft of labor as well. In a Motion for Summary Judgment filed on September 17, 2026, by the New York Times in its ongoing case New York Times v OpenAI/Microsoft[19] the Times quoted their opponents extensively to make this exact point – again, if you pay attention to the thieves they’ll happily tell you what they’re doing. Microsoft’s Director of Applied Science referred to AI as “an astonishing theft of unprecedented proportions,” and “the largest theft of labor in human history.” OpenAI’s Head of ChatGPT wrote that “publishers face an ‘existential threat’ from those products, SF1466” which, he said, “are largely substitutive, period,” and “will get more and more substitutive as they get better.” Substitutive, if you’re not familiar with the term, means that the new product entirely replaces and displaces the original – a clear violation of copyright law. “Defendants repeatedly copied millions of the Plaintiffs’ copyrighted articles in their entirety without permission to produce substitutive commercial AI products,” noted the Times. “Such admissions eviscerate Defendants’ ‘fair use’ defense because substitution is ‘copyright’s bête noire.’ Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508, 528 (2023).”

In a striking example of AI’s theft of both intellectual property and labor, last month OpenAI breathlessly announced that its AI had solved the long-standing Navier-Stokes problem after only 88 hours of work. But the work for this was done by Tristan Buckmaster and Levent Alpöge, two mathematicians who had been working on this problem for years by this point. OpenAI explicitly admitted that they can’t rule out having taken Buckmaster and Alpöge’s data and used it for their own models without consent, credit, or compensation, and they saw nothing wrong with that. The more you read about this case, the more it becomes the simple garden-variety theft of labor that all AI is. It’s an ongoing case, but when you’ve managed to rile up mathematicians to action you’ve screwed up pretty badly.

Another example of this theft comes from Alexandra Tremayne-Pengelle’s article in last month’s Atlantic Magazine.[20] Apparently AI models are not satisfied with just stealing the copyrighted works of authors without permission, credit, or compensation to sell it for their own gain anymore. People are now using AI to generate new works attributed to those authors, extrapolated from that original theft into new manuscripts, stealing the authors’ hard-earned reputations and accumulated goodwill. This is also common in the music industry these days, according to Tremayne-Pengelle. Folks, this is how you stop people from making music or writing books, and maybe that’s okay with the sort of people who run AI companies but it can’t be acceptable to anyone with an ounce of moral fiber.

The entire business model of AI is based on stealing from people who do the actual work without giving them anything in return, and if you think there is anything even remotely ethical about that you need to re-examine your life choices. 

------


[1] Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task (Nataliya Kosmyna, Eugene Hauptmann, Ye Tong Yuan, Jessica Situ, Xian-Hao Liao, Ashley Vivian Beresnitzky, Iris Braunstein, & Pattie Maes)


[2] The Cognitive Divergence: AI Context Windows, Human Attention Decline, and the Delegation Feedback Loop (Natanel Eliav (Machine Human Intelligence Lab)


[3] Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians (KartiK Chandra, Max Kleiman-Weiner, Jonathan Ragan-Kelley, & Joshua B. Tenenbaum, scientists employed by MIT and the University of Washington-Seattle)


[4] An Echo Chamber of One: Should AI Psychosis Be a Distinct Clinical Entity? (Joshua Au Yeung, Hamilton Morrin, Vincent Ng, Zeljko Kraljevic, & Richard Dobson, August 26, 2026)


[5] AI Empathy Erodes Cognitive Autonomy in Younger Users (Junfeng Jiao, Abhejay Murali, Saleh Afroogh, Urban Information Lab, Austin Texas, April 1 2026)


[6] “I wish I could push ChatGPT Off a Cliff:” Professors scramble to save critical thinking in an age of AI (The Guardian, March 10, 2026)


[7] A New Direction for Students in an AI World: Prosper, Prepare, Protect (Mary Burns, Rebecca Winthrop, Natasha Luther, Emma Venetis, & Rida Karim)


[8] Generative AI without guardrails can harm learning: evidence from high school mathematics (Hamsa Bastani, Osbert Bastani, Alp Sungu, Haosen Ge, Özge Kabakci, & Rei Mariman)


[9] US datacenter discharged large amounts of pollution for almost a year with no permits. (Evan Simon, The Guardian, October 1, 2026)


[10] “’Whata Bod’: An AI-generated NWS map invented fake towns in Idaho.” (Washington Post, January 6, 2026)


[11] Moffat v. Air Canada, 2024 BCCRT 149 (CanLII)


[12] Mobley v Workday, US District Court, Northern District of California, Case# 23-cv-00770, June 22, 2026 order from Judge Rita F. Lin


[13] The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity (Parshin Shojaee, Iman Mirzadeh, Keivan Alizadeh, Maxwell Horton, Samy Bengio, & Mehrdad Farajtabar)


[14] AI Search Has a Citation Problem (Klaudia Jaźwińska & Aisvarya Chandrasekar, Columbia Journalism Review, 3/6/2025)


[15] Fabricated citations in the age of AI: A wake-up call for editors, reviewers, and authors (Journal of Dental Sciences 21 (2026) 679-680)


[16] LLMs Corrupt Your Documents When You Delegate (Philippe Laban, Tobias Schnabel, & Jennifer Neville; Microsoft Research April 17, 2026)


[17] AI models collapse when trained on recursively generated data (Ilia Shumailov, Zakhar Shumaylov, Yiren Zhao, Nicolas Papernot, Ross Anderson, & Yarin Gao) Nature 631, 755-759 (2024)


[18] OpenAI – Written Evidence (LLM0013), submitted 5 December 2023


[19]In Re: OpenAI, Inc., Copyright Infringement Litigation, Case Nos. 1:23-cv-11195-SHS-OTW, 1:24-cv-01515-SHS-OTW, 1:24-cv-03825-SHS-OTW, 1:24-cv-04872-SHS-OTW, 1:25-cv-04315-SHS-OTW; collected as 25-md-3143 (SHS) (OTW)


[20] The Rise of Parasite Authors (The Atlantic, September 18, 2026)