Brian Merchant

Responding to a widely covered resignation statement that lab staff fear AI could kill everyone, he tells readers not to worry that AI will go rogue and kill us all: he has found no credible step-by-step account from self-improving AI to human extinction. Extinction declarations are a kind of marketing, but he says anyone calling them only marketing is as wrong as those who deny the marketing. The real danger is companies with concentrated compute, surveillance capacity and military and state ties experimenting recklessly without consequences. He wants full model transparency, liability and movement toward public control, with dismantling the industry on the table pending democratic input, and opposes an embedded AI-safety-derived proposal as likely regulatory capture. Full post inspected.

Argues that the labs’ vocabulary (“extinction risk,” “p(doom),” “AGI,” “going rogue”) serves their narrative interests the way Trump’s “super intelligence” rebranding serves his. He lists three benefits to the labs: investor value before IPOs, regulatory capture through industry-friendly standards and audits, and, following Matt Levine’s argument, a legal pretext for a coordinated slowdown. Calling incidents “going rogue” absolves companies, so he worries more about company ethics and malign actors than about technology acquiring a life of its own. Arguments from writers he links and the second-half book excerpt by other authors are not his. First part inspected in full.

His opening monologue asks why companies whose staff say their technology might kill everyone keep building it. He calls their justifications (beating China, prosperity worth the risk) vague and sociopathic, and says anyone who believed in a one-in-ten chance of killing billions would stop. Those who believe and continue are, in his view, sociopaths or people who think only they should command such power; others may not believe it and are going along for wealth and power. He concludes that people with that moral calculus should be stopped from concentrating power. This reasons from the labs’ premises; it is not his own extinction estimate. Published monologue transcript inspected; guest conversation audio not reviewed.

Case study of Anthropic announcing a model too dangerous to release, receiving extensive press, closing a very large funding round and then selling a restricted “Mythos-class” model two months later. He defines doom marketing as drumming up investor and media interest by making harrowing claims about threats to jobs, norms or humanity’s existence, and says the media fell for it while cybersecurity professionals’ criticisms were sidelined. The case is about one campaign’s framing; he relies on others’ technical critiques and does not establish what the model could do. Opening sections inspected; later unrelated items not used.

Using a Glassdoor report (others’ data), he argues that how workers feel about AI depends on how much power they have at work: executives are positive, while accountable frontline roles such as claims adjusters, accountants and IT workers are negative. From his own worker interviews he finds it more common for AI policies to make workers miserable than to eliminate their jobs. He ties executives’ jobs-apocalypse proclamations to selling automation to management and links workplace resentment to backlash against data centers, surveillance cameras and AI glasses. Full post inspected.

Says the executives’ forecast jobs apocalypse has arrived only “kind of”: instead of replacing tens of millions of workers, AI is mostly making them miserable. He praises New York City’s ban on AI in K–8 classrooms, wondering why it was not made permanent, and treats California bills curbing AI and social media harms as productive channels for worker anger. Quoted reporting and the interview audio are not his views and were not used. Written portions inspected.

Argues that data centers, license-plate surveillance cameras and AI glasses were imposed on public life with little democratic input and that people are rejecting them because of what they do and represent, sharpened by inequality and a sense of powerlessness. He mocks industry responses that blame psy-ops or psychosis. This is an analysis of public sentiment and a normative claim about democratic consent, not a capability forecast. Full post inspected.

Older context. His critique of a tech podcast argues that adopting the industry’s AGI framing amplifies a sales pitch for automation and lets executives off the hook for human decisions. The Luddites fought factory owners using machines to cut wages and deskill them, not technology itself, and industrial automation degraded rather than abolished cloth workers, a pattern he fears for creative workers. Who benefits from automation is a question of power, and mass-job-loss prophecies have historically come from elites. Hosts’ responses are excluded. Full post inspected.