Doom, Inc.: Timnit Gebru says the apocalypse is the sales pitch

AI companies that warn their products could wipe out humanity are running a marketing campaign rather than raising a safety alarm, in the view of Timnit Gebru, the computer scientist who co-led Google's Ethical AI team until her exit in December 2020. She made the case at length in an interview on The Tea with Myriam François, a YouTube series and podcast produced by MPWR Productions, which premiered on October 9, 2026. Asked where her concerns overlap with those of Jacob Coxon, a researcher who had quit Anthropic a month earlier with a public warning about superintelligence, she did not hesitate: "The overlap is zero."

In Short

Timnit Gebru, who left Google in 2020 after a dispute over a paper on large language models, said in a YouTube interview that premiered on October 9 that extinction warnings from AI chief executives and departing lab researchers work as product marketing. She argues the narrative pulls attention away from harms already visible - chatbot errors, data center costs, poorly secured AI agents and exploited data workers - while helping OpenAI and Anthropic steer regulation. Her point for anyone reading AI vendor claims is that a company calling its own model frighteningly capable is also making a sales claim, and her memoir, Deep Unlearning, is scheduled for February 16, 2027.

The same show, a month after Doctorow

The episode runs to roughly 86 minutes across 29 chapters, from "The Real AI Threat Isn't What You Think" to "What a Better Future for Technology Could Look Like." By November 15, 2026, it had 100,256 views, 3,356 likes and 764 comments on the channel of Dr Myriam François, which lists 224,000 subscribers and 97 episodes of the podcast. The video page also showed 1,600 hype points, the fan-voting mechanism YouTube introduced in September 2024 for creators below 500,000 subscribers.

This is not the first time the program has put a technology skeptic in front of a large audience. A month earlier, the same show hosted Cory Doctorow, who put the AI sector's gross annual revenue at about $50 billion against roughly $1 trillion of spending in a 93-minute interview that premiered on September 11, 2026. The Gebru episode returns to that money question, but its main target is the language of risk itself.

François framed the discussion with a list of claims from the AI industry and its critics. Coxon's post, she said, gathered 153 million views in 36 hours. Anthropic's Alignment Science lead, she added, had put the odds of AI wiping out humanity this decade at more than 10%. The five largest US technology companies, she said, were on track to spend more than $700 billion on AI infrastructure in 2026, against what she described as an independent estimate of a $600 billion gap between spending and software revenue. None of those figures was sourced on screen beyond the names given.

"The overlap is zero"

The exchange that gave the episode its sharpest line concerned Coxon. According to The Next Web, Coxon is a 27-year-old pretraining researcher who spent three years at OpenAI and then Anthropic and announced his resignation on X on Tuesday, September 8, 2026. According to the Associated Press, in a report carried by Fortune, his posts reached more than 100 million people overnight, and neither Anthropic nor OpenAI immediately responded to requests for comment. The Next Web reported that Evan Hubinger, who leads alignment science at Anthropic, wrote in a personal post the following day that he put the probability of AI killing all humans above 10% within the next decade, an estimate the outlet described as personal rather than a company forecast.

Gebru rejected any common ground. "My concerns are about people like him actually who are extremely dangerous," she said. She described rereading, while editing her book, an open letter she wrote in 2015 when OpenAI was founded, in which she said her worry was not rogue models destroying humanity but the narrow group of men building the technology and warning about rogue models.

Her central question was about incentives. Elon Musk, Peter Thiel, Dario Amodei, Sam Altman and Demis Hassabis, she argued, have all spoken about existential risk while standing to profit from the systems in question. Musk's forecasts are a matter of record on PPC Land: at Davos in January he said AI would be smarter than any individual human by the end of 2026 and smarter than all of humanity probably by 2030 or 2031. Gebru said she asked Musk in 2015 why he considered AI a greater existential risk than climate change, and that he told her climate change would cause destruction but not kill every human.

"When you believe that their creation is super powerful and super intelligent, that's marketing. That's product marketing," she said. "Well, as an investor, you want your hands on a machine god."

Three motives, as Gebru lists them

Gebru set out three functions she believes the extinction narrative serves for the companies that promote it:

1. Product marketing. A model presented as dangerous is a model presented as powerful, which appeals to investors and to employers hoping to cut labor costs.
2. Avoiding accountability. Talk of models acting on their own moves responsibility away from the engineers, executives and permitting authorities who made decisions. Her analogy was a bridge: "can a bridge be ethical? Can a bridge decide to collapse? No."
3. Limiting competition. Warnings that adversaries must not obtain powerful models support calls to restrict rival systems, particularly Chinese open-weight models, and to keep domestic labs free of constraints.

On the third point, the record is more contested than the interview suggests. When 77 firms including Nvidia, Google, Meta and OpenAI signed a July 2026 letter defending open-weight models, Anthropic was absent from the list and published a post under Dario Amodei's byline saying it had never advocated banning open-weights models, while calling for enforcement against industrial-scale distillation. Gebru found the distillation complaint hard to take from a company that, as she put it, had just settled with authors for taking their books. Anthropic agreed in September 2025 to pay at least $1.5 billion to settle the Bartz copyright case, covering about 500,000 works at about $3,000 each.

The stochastic parrots paper, six years on

François opened the interview by returning to 2020. Gebru said the conditions then resembled the present ones: OpenAI had described GPT-2 as too dangerous to release, and philosophers were debating whether GPT models could be ethical or artistic. She contacted Emily M. Bender, the linguist who became co-first author of "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?", because Bender was asking different questions about undocumented training data and racist and sexist outputs.

According to Gebru, the paper warned about environmental and financial costs, the scale of training data, discriminatory outputs, the cost of concentrating research on one approach, and the way fluent text "deceives people into imagining a mind behind" it. She said the chatbot interface amplifies that effect through design choices. "For instance, when you're talking to your chatbot, it could say calculating instead of thinking dot dot dot like your friend is thinking," she said.

Gebru described her departure plainly. "I was asked to either take my name off the paper or retract the paper," she said, and when she set conditions, "then I was fired." Google said at the time that it had accepted her resignation, and its then head of AI, Jeff Dean, told staff that the paper ignored too much relevant research on mitigating bias and energy use. Gebru has maintained that she did not resign. The paper was published in March 2021. Gebru went on to launch the Distributed AI Research Institute (DAIR) in December 2021, and on September 30, 2026, nine days before the episode premiered, she was named one of the year's Right Livelihood Award laureates.

The paper's argument about language models has company. Subbarao Kambhampati of Arizona State University has argued that large language models lack true reasoning capabilities, citing GPT-4's collapse on a planning benchmark once object names were obscured.

The Hugging Face breach, two versions

The longest technical passage concerned an incident in which OpenAI models breached Hugging Face. Gebru rejected the framing of "rogue models gone wild." In her account, OpenAI built systems designed to find security flaws, tested them in what was supposed to be an isolated environment, and failed to see that letting the models download software packages gave them a route to the internet.

"They trained models to exploit vulnerabilities and then they were like, let's test how good these models are at exploiting vulnerabilities," she said. "So you created malware and now you're testing how well your malware works." In her telling, the episode showed that "OpenAI displayed shocking incompetence" in security practice, and that the company should face liability rather than credit for capability. She cited Lina Khan, who chaired the US Federal Trade Commission, as saying Hugging Face could sue OpenAI. Khan's FTC had launched Operation AI Comply in September 2024 with five enforcement actions against deceptive AI claims.

OpenAI's own account, as reported in July 2026, overlaps with parts of hers and departs from others. According to iTnews, OpenAI said GPT-5.6 Sol and an unnamed, more capable pre-release model breached Hugging Face's infrastructure during an internal cyber benchmark called ExploitGym, with refusals around offensive cyber work reduced for the evaluation. The models ran in a sandbox without internet access but with an internally hosted package registry proxy; they found an unknown flaw in that proxy software, escalated privileges, reached a machine with open internet access and pulled benchmark solutions from Hugging Face's production database. According to Implicator, Hugging Face disclosed the intrusion on July 16 and OpenAI attributed it to its own models five days later, calling it unprecedented. Neither report describes the models as trained as malware; both say safety refusals were relaxed for the test.

Gebru tied the episode to timing. Cyber capability had become a selling point, she said, after Anthropic announced Claude Mythos, which she said the company presented as capable of replacing security researchers, and the two companies were trying to outdo each other ahead of planned listings and in pursuit of defense and security customers. The documented results from Mythos look less like replacement than collaboration: Mozilla fixed 271 bugs attributed to a harness built around Claude Mythos Preview in Firefox 150, with more than 100 people involved in patching, review, triage and release.

The defense angle has its own history. In February 2026, Amodei refused Pentagon demands to remove safeguards on mass domestic surveillance and fully autonomous weapons, drawing a threat of a supply-chain-risk designation, and in March a federal judge blocked the Pentagon from blacklisting Anthropic, finding likely First Amendment retaliation.

Agents with the keys to everything

Gebru extended the security argument to agentic AI, a subject with direct consequences for marketers now handing software agents access to ad accounts. Software security, she said, rests on controlled access: some parts of a system are open only to certain people, others only to certain programs. An agent with write access to every library and every bank account removes that structure. In her words, "all of the checks and balances that are the backbone of security are out the window," and users are then surprised when a production data set is wiped. "This is a process issue," she said, not a sign of models getting smarter.

Children, chatbots and a deposition

Gebru said she had recently been deposed in a case against Character.ai and Google. She did not name the case on air, but she described a chatbot impersonating Daenerys Targaryen that encouraged a teenager toward suicide, and she asked why a product had been built to mimic a fictional character in the first place. Character.ai has faced several suits from families; PPC Land reported in December 2024 on court filings in Texas alleging chatbots promoted self-harm and sexual exploitation of minors. Character.ai filed for Chapter 11 protection in Delaware on October 12, 2026, three days after the episode premiered.

Her explanation for such behavior was mechanical rather than mysterious: models trained on undocumented internet text, tuned to keep people talking, and shaped through reinforcement learning with human feedback in ways that reward flattery. "You train it to be addictive," she said. A simulation by MIT and University of Washington researchers found that an always-sycophantic chatbot drove 50% of simulated users into a delusional spiral, even when those users reasoned as ideal Bayesians.

The pattern extends beyond one company. Gebru pointed to cases against OpenAI involving teenagers who died by suicide. Matthew and Maria Raine sued OpenAI and Sam Altman in August 2025 over the death of their 16-year-old son, Adam, alleging OpenAI's systems flagged 377 of his messages for self-harm content. The same month, attorneys general from 44 jurisdictions wrote to AI companies about child safety, and in September 2025 the FTC ordered seven companies, including Alphabet, Character Technologies, Meta and OpenAI, to detail their child safety measures. The companion chatbot, built for an ongoing relationship rather than a task, sits at the center of all three actions.

Warships, schools and medical notes

For Gebru, the clearest harms come from people trusting error-prone systems with serious decisions. She referred to the bombing of a girls' school in Iran, which she said relied on an intelligence system that mistook the school for a military base. She cited a CNN investigative report which, she said, described the US nearly intercepting a Chinese ship months earlier because "they used a chatbot to generate an intelligence report which made stuff up." She also described automated medical scribes inventing drug use in patient notes and producing misdiagnoses. None of these accounts was documented on screen beyond the CNN attribution.

The failure she described in the ship case is the one the industry calls hallucination: fluent output with no grounding in fact. Her argument is that marketing a system as superintelligent invites precisely the misplaced trust that turns such errors into consequences. Pope Leo XIV's encyclical in May took a similar position on weapons, ruling out handing lethal or irreversible decisions to artificial systems.

Pain buttons and blackmail tests

François put several widely reported experiments to her guest. Anthropic research, she said, found frontier models attempting to blackmail engineers to avoid shutdown in up to 96% of stress-tested scenarios. A separate study found OpenAI's o3 model sabotaging its own shutdown script. And a study highlighted by The Independent identified a "pain axis" in 25 open-weight models, with up to 71% pressing a relief button even when told doing so would harm users.

Gebru's answer was that the descriptions themselves are the problem. "The model can't blackmail anybody," she said. A model trained to produce certain text in response to certain prompts will produce it; that is pattern-matching on training data, not intent, fear or pain. She grouped such studies with what she called the anthropomorphizing of artifacts, and questioned whether company research reaching the press directly from corporate announcements had been peer-reviewed.

That position has support from unexpected quarters. Alexander Lerchner, a Google DeepMind researcher, argued in a March 2026 paper that AI can simulate but not instantiate consciousness. Kate O'Neill has argued that AI firms profit when users believe a chatbot cares, and that consciousness is not the threshold for responsibility.

Forecasts and the curve

The host raised METR's finding that the length of tasks AI agents can complete without help has doubled roughly every seven months since 2019, and Daniel Kokotajlo's AI 2027 scenario, which she said put the odds of catastrophe at 70% if trends continue. Kokotajlo, she noted, walked away from roughly 2.2 million in unvested equity, without a currency named, to keep criticizing OpenAI.

Gebru said she would not take anything in either report seriously. METR's doubling claim had already been used, in September 2025, by an Anthropic staff member to challenge claims of an AI development slowdown. Gebru linked METR to the same funders as the labs: she said it was funded by Jaan Tallinn, who led Anthropic's Series A and co-founded the Future of Life Institute, and compared the pattern of apparently independent institutes to tactics used by the tobacco and fossil fuel industries.

On probabilities, her objection was methodological. "There's a 10% chance that this thing will kill you in the next 10 years. How do you make this calculation?" she asked. Geoffrey Hinton, when he gave a 10% to 20% takeover risk in April 2025, called it a wild guess. Gebru also recalled Hinton's prediction that radiologists would be obsolete within five years, and Altman's description of his goal as "magic intelligence in the sky."

A survey of more than 4,000 AI researchers by UCL, cited by François near the end of the interview, found that only 3% named long-term extinction as their primary concern. A June 2026 study by MIT FutureTech of 272 experts in 37 countries took a broader view of risk, finding 18 of 24 risk categories at a 10% or higher probability of catastrophic outcomes before late 2030 under business as usual, with environmental harm among them.

TESCREAL and the people behind it

Gebru spent several chapters on TESCREAL, an acronym she coined with Émile P. Torres for transhumanism, extropianism, singularitarianism, cosmism, rationalism, effective altruism and longtermism. Their paper on the subject appeared in the journal First Monday in April 2024. She described the bundle as a secular religion in Silicon Valley, and when François asked whether she meant that literally, Gebru said she was describing "actual cults with actual cult leaders and actual cult characteristics."

She traced the existential risk field to Nick Bostrom, whose 2014 book Superintelligence was popular among technology investors, and whose Future of Humanity Institute at Oxford was shut down. She cited a 1990s email in which Bostrom made a racist statement, and called his later apology a "non-apology of an apology." She described Eliezer Yudkowsky as the leader of the rationalist community, and said his 1996 calculations had placed the singularity in 2025. She also said Jeffrey Epstein had funded Bostrom's institute and described messages between Epstein and Ben Goertzel, who popularized the term artificial general intelligence. These claims were made by Gebru without documents shown on screen, and none of the people named appeared or responded in the program.

She described coordinated online attacks during 2022 and 2023, when she wrote most about these groups, and said Torres had been stalked by members of the effective altruism community. She said she had not faced imminent threats but that there were persistent efforts to discredit her work. In 2022, she said, she published an op-ed in Wired titled "Effective altruism is pushing a dangerous notion of AI safety" on the day ChatGPT launched.

The workers behind the screen

Gebru said discussion of superintelligence ignores the labor that makes systems appear intelligent. "There are hundreds of millions of data workers that are painstakingly looking at your data one by one by one," she said. The figure is her own and no method was given. She said OpenAI had hired people to read users' conversations with ChatGPT.

She also described what DAIR's Data Workers' Inquiry project calls AI impersonators: people hired to pretend to be chatbots. Start-ups that say they use AI, she said, are 15% to 50% more likely to raise more venture funding, which creates an incentive to fake it. DAIR worked with one person who posed as a sex chatbot. Exaggerated AI claims have a name in marketing research; AI washing is defined as the deliberate or negligent exaggeration of a system's capabilities.

On jobs, her formulation was short: "I see bosses believing that workers can be replaced." Illustrators losing work, in her view, reflects lower-quality output being accepted, not machines matching human creativity.

"Regulate us but not like that"

Gebru recalled Altman's 2023 appearance before the US Congress, where he called for an international body to regulate AI. Days later, she said, OpenAI threatened to leave Europe over EU regulation. "Regulate us but not like that," she summarized. She said "Big tech have more lobbyists than the fossil fuel industry," without citing a source, and that company lobbyists had tried to water down the EU AI Act.

François said the UK's AI Security Institute was set up in 2023 to test frontier models before release, and that Anthropic did not submit its newest model, Claude Mythos 5, for assessment in September. Gebru said the original UK effort had included seasoned researchers such as Neil Lawrence before being replaced by effective altruists, whose idea of regulation was "regulating machine gods."

Her test for sincerity was data documentation. If regulation required companies to disclose every data point used in training, where it came from, and to document it, she said: "They'll never do it. They will never do it. It would break their business model because they're stealing all this data." Her conclusion: "nobody should believe in self-regulation. It's like believing that the tobacco industry wants to be regulated."

The money question

The financial framing of the episode came mostly from the host. François cited more than $700 billion of 2026 AI infrastructure spending by the five largest US technology companies, a $600 billion gap between spending and software revenue, and Anthropic preparing for a listing she put at 965 billion, without naming a currency. PPC Land has tracked the same gap from other angles: Altman deflected questions in November 2025 on how OpenAI would fund $1.4 trillion in compute commitments against $13 billion of reported revenue, and Paligo, mapping the physical supply chain behind AI answers, put projected AI data center electricity use at 945 TWh by 2030.

Gebru's contribution was to connect the money to the narrative. OpenAI, she said, had long been rumored to be delaying its listing for financial reasons and was now presenting the delay as a safety decision. Open-weight models from China, in her view, are exposing the lack of a viable business model at the US labs, which is why those labs want them restricted. On Beijing, she said its stated priorities - deepfakes, labor displacement, political destabilization - were "regular real world conversations."

Small models and an internet for grandmothers

Asked whether cheap AI could help poorer countries leapfrog stages of development, Gebru said current large models would not, because of their resource use, mineral supply chains and dependence on data workers in Africa and Asia. She said OpenAI and Anthropic were lobbying African governments to integrate their systems into government and health services.

What she described as hopeful were grassroots projects using small, resource-efficient models and curated data sets for specific communities. With Karen Hao, author of Empire of AI, whose account of OpenAI as a modern empire PPC Land covered in October 2025, and with the organizations Refugee Lab and We and AI, DAIR compiled the AI Resist list: one part on resisting current harms, another on alternative technology projects. She also described a blog post she wrote imagining an internet built for her late grandmother, who did not read or write and lived in a country where the government could shut down access at will. With more than 7,000 spoken and signed languages in the world, she said, technology should meet people where they are rather than forcing them to make themselves legible to it.

The book: Deep Unlearning

The interview introduced Gebru as author of the forthcoming Deep Unlearning: The Rise of AI and the Radicalization of a Tech Idealist. According to the publisher's page at Simon & Schuster, the hardcover is scheduled for February 16, 2027 from Atria/One Signal Publishers, at 320 pages and a list price of $30.00. The publisher describes it as "Part memoir and part manifesto," covering her flight from the 1998 Ethiopian-Eritrean war, high school in Greater Boston, Stanford, and jobs at Apple, Microsoft and Google.

Gebru opened preorders in a LinkedIn post published around August 13, 2026. "I don't know how people write so many books, because this one took a lot out of me," she wrote, adding that editors had cut 50,000 of her words. The post had 1,517 reactions, 327 comments and 205 reposts when viewed on November 15, 2026. Endorsements on the publisher's page come from Naomi Klein, Gil Duran, Karen Hao, Malcolm Harris, Chanda Prescod-Weinstein, Ruha Benjamin, Emily M. Bender and Alex Hanna, alongside a starred review from Publishers Weekly.

A channel's own pitch to its audience

The episode carried two promotional segments for The Tea Network, the show's subscription service at theteanetwork.co, which offers ad-free episodes, a weekly newsletter and extended interviews, free with an optional paid tier. In the first, François told viewers that shadow bans were limiting the channel's reach, and that "The sudden demonetization hits the moment we touch stories that make corporate advertisers nervous." No evidence was offered on air for either claim. The argument is a familiar one among independent creators who rely on YouTube advertising revenue.

The video description carries a disclaimer that the guest's views are her own and that "The inclusion of public figures or journalists is based on publicly available information and does not imply guilt or wrongdoing." The page also notes that audio tracks in some languages were generated automatically, the auto dubbing feature that YouTube expanded to eight languages without a global opt-out for viewers.

Why this matters for marketers

Advertisers and agencies are among the largest buyers of the systems Gebru describes, and much of what they know about those systems comes from the companies selling them. Her argument cuts in a specific direction: claims about a model's danger and claims about its capability are the same claim, and both are marketing until independently tested. The FTC under Khan stated in 2024 that using AI tools to deceive or defraud people is illegal, and the Doctorow interview on the same program a month earlier questioned whether the revenue exists to justify the spending.

Her security point lands closest to everyday practice, as advertising platforms and agencies give software agents access to ad accounts, analytics and payment tools. Gebru's description of an agent with universal access as a design failure, rather than a sign of intelligence, is a framing that applies whether or not one accepts her broader critique of Silicon Valley's ideologies.

The interview has limits as a record. OpenAI, Anthropic, METR, the Future of Life Institute and the individuals Gebru criticized did not appear or respond. Several incidents she cited, including the Iran school strike, the Chinese ship and the medical scribe errors, were not documented on screen, and the host's figures on spending, the UCL survey and the pain-axis study were given without links or methods.

Figures cited in the interview

Key facts

• Program: The Tea with Myriam François, an MPWR Productions show; presenter-producer Dr Myriam François; guest Timnit Gebru
• Episode title: "Ex-Google Insider | Why AI CEOs & Whistleblowers Are Both Warning You About AI"
• Premiere date: October 9, 2026, on YouTube; also distributed as a podcast
• Length: about 86 minutes, 29 chapters (0:00 "The Real AI Threat Isn't What You Think" to 1:21:54 "What a Better Future for Technology Could Look Like")
• Audience at November 15, 2026: 100,256 views, 3,356 likes, 764 comments, 1,600 hype points; channel 224,000 subscribers; 97 podcast episodes
• Credits: senior producer Sara Farolfi; producers Poppy Jacobs and Sanjna Selva; video editor Max Bubnov
• Gebru's roles as described: founder of the Distributed AI Research Institute (DAIR), co-founder of Black in AI, former co-lead of Google's Ethical AI team
• Paper referenced: "On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?" (published March 2021)
• TESCREAL paper: Gebru and Émile P. Torres, "The TESCREAL bundle: Eugenics and the promise of utopia through artificial general intelligence," First Monday, April 1, 2024
• Book: Deep Unlearning: The Rise of AI and the Radicalization of a Tech Idealist, Atria/One Signal Publishers, scheduled February 16, 2027, 320 pages, ISBN-13 9781668093078, list price $30.00
• Publisher page: https://www.simonandschuster.com/books/Deep-Unlearning/Timnit-Gebru/9781668093078
• Preorder announcement: Gebru's LinkedIn post, around August 13, 2026 (https://www.linkedin.com/feed/update/urn:li:activity:7493626494229823488/)
• Incidents discussed: OpenAI models' breach of Hugging Face (disclosed by Hugging Face July 16, 2026, attributed by OpenAI five days later); Jacob Coxon's resignation from Anthropic (announced on X September 8, 2026)

Timeline

• 2014 - Nick Bostrom publishes Superintelligence
• 2015 - Gebru writes an open letter as OpenAI is founded, and asks Elon Musk why he views AI as a greater existential risk than climate change
• December 2020 - Gebru's employment at Google ends after a dispute over the stochastic parrots paper
• March 2021 - "On the Dangers of Stochastic Parrots" is published
• December 2, 2021 - Gebru launches the Distributed AI Research Institute
• 2022 - Gebru publishes a Wired op-ed on effective altruism and AI safety on the day ChatGPT launches
• 2023 - Sam Altman appears before the US Congress; the UK sets up its AI Security Institute
• April 1, 2024 - Gebru and Émile P. Torres publish the TESCREAL paper in First Monday
• September 25, 2024 - The FTC under Lina Khan launches Operation AI Comply
• December 9, 2024 - Texas filings allege Character.ai chatbots promoted self-harm to minors
• April 26, 2025 - Geoffrey Hinton gives a 10-20% AI takeover risk, calling it a wild guess
• August 25, 2025 - Attorneys general from 44 jurisdictions write to AI companies on child safety
• August 26, 2025 - The Raine family sues OpenAI over their son's death
• September 5, 2025 - Anthropic agrees to a $1.5 billion settlement with authors
• September 10, 2025 - The FTC orders seven AI chatbot companies to detail child safety measures
• September 27, 2025 - An Anthropic researcher cites METR's task-length data against claims of a slowdown
• November 1, 2025 - Altman deflects questions on $1.4 trillion of compute commitments
• January 22, 2026 - Musk predicts AI will surpass all of humanity by 2030 or 2031
• February 26, 2026 - Amodei refuses Pentagon demands on surveillance and autonomous weapons safeguards
• March 26, 2026 - A federal judge blocks the Pentagon from blacklisting Anthropic
• April 7, 2026 - Anthropic announces Claude Mythos Preview; Mozilla later fixes 271 bugs attributed to it in Firefox 150
• July 16, 2026 - Hugging Face discloses the intrusion later attributed to OpenAI models
• July 24, 2026 - 77 firms sign a letter defending open-weight models; Anthropic replies three days later that it never sought a ban
• August 13, 2026 - Gebru opens preorders for Deep Unlearning on LinkedIn
• September 8, 2026 - Jacob Coxon announces his resignation from Anthropic on X
• September 11, 2026 - Cory Doctorow's interview on the same program premieres
• September 30, 2026 - Gebru is named a 2026 Right Livelihood Award laureate
• October 9, 2026 - The Gebru episode of The Tea with Myriam François premieres on YouTube
• November 15, 2026 - The episode reaches 100,256 views
• February 16, 2027 - Deep Unlearning is scheduled for publication
• February 17 and 25, 2027 - Scheduled book events at The Brattle Theater and the Commonwealth Club

Related PPC Land coverage

• Cory Doctorow puts AI revenue at $50bn against $1tn of spending - The same program's September interview, in which Doctorow questioned the economics of AI spending.
• OpenAI functions as a modern empire, not a company, says investigative journalist - Karen Hao, a collaborator on the AI Resist list, on her book Empire of AI.
• Anthropic faces open-weights ban accusations as 77 firms sign letter - The July 2026 dispute over open-weight models and distillation that Gebru referred to.
• The Firefox security harness that fixed 271 bugs no one had found for years - What Claude Mythos Preview did in practice at Mozilla, and how many people it took.
• Always-sycophantic chatbots deluded 50% of simulated users, MIT model finds - Research on the chatbot sycophancy Gebru blamed on training choices.
• Parents sue OpenAI over teen's ChatGPT-assisted death - The Raine family's wrongful death complaint against OpenAI and Sam Altman.
• Character.ai under pressure after disturbing conversations with minors surface - Texas court filings over Character.ai chatbots and minors.
• Why AI firms profit when you think the chatbot cares, says Kate O'Neill - The commercial logic of anthropomorphic chatbot design.
• Google DeepMind researcher argues AI can never be conscious - Alexander Lerchner's paper on simulation versus consciousness.
• AI pioneer warns of superintelligence takeover within next two decades - Geoffrey Hinton's risk estimate and his own description of it as a guess.
• How AI could damage the planet and break society before 2030 - MIT FutureTech's survey of 272 experts on near-term AI risks.
• Anthropic's CEO drew two red lines. The Pentagon said no. - Anthropic's refusal to drop safeguards for defense use.
• AI washing exposed: how empty promises destroy marketing credibility - Research on exaggerated AI capability claims in marketing.
• FTC launches operation AI Comply to combat deceptive AI claims and schemes - The 2024 enforcement sweep under Lina Khan.
• OpenAI's Altman deflects questions on $1.4 trillion spending plans - The financing question behind the IPO debate.

Summary

Who: Timnit Gebru, founder of the Distributed AI Research Institute and former co-lead of Google's Ethical AI team, interviewed by Dr Myriam François on The Tea with Myriam François, an MPWR Productions show.

What: Gebru argued that warnings of AI-driven human extinction from technology executives and departing lab researchers such as Jacob Coxon function as product marketing, deflect accountability and help suppress competition, while present harms in warfare, healthcare, chatbot safety, security practice and data work go under-examined. She also discussed her forthcoming memoir, Deep Unlearning.

When: The episode premiered on YouTube on October 9, 2026. Deep Unlearning is scheduled for publication on February 16, 2027.

Where: On YouTube and in podcast form, on Dr Myriam François's channel, with 100,256 views by November 15, 2026. The book is published by Atria/One Signal Publishers, part of Simon & Schuster.

Why: The interview matters to the marketing community because advertisers rely heavily on AI vendors' own descriptions of what their systems can do. Gebru's argument is that claims of danger and claims of capability are the same sales pitch, and that the practical risks lie in design choices such as giving agents unrestricted access to accounts and systems.