AI Apocalypse Isn't Happening. This is Worse...

An interview recorded days after Jacob Coxon resigned from Anthropic on 8 September 2026 (Time), warning that the labs are racing toward self-improving superintelligence. Anthropic’s alignment science lead, Evan Hubinger, backed him on X, putting the chance that AI kills all humans at more than 10% within the next decade. Doctorow reads both as criti-hype, Lee Vinsel’s 2021 term for criticism that inflates the hype it sets out to attack. A system with a one-in-ten chance of ending the species must surely be able to do your job, so the doom story doubles as the replacement pitch. In Doctorow’s paraphrase, the pitch ends at “we can fire the last radiologist.” He times it to an Anthropic IPO he puts three weeks out. That matched reporting rather than any announcement: Business Insider reported an October Nasdaq target, and Anthropic had confirmed no date (The Next Web, 13 September).

His claim is that AI’s real use is disciplining labour, as the machines disciplined the Luddites. From Riley Quinn he takes the idea that automation did not eliminate weavers, it eliminated high-quality cloth, while carpenters, whose work could not be automated, still lost ground to the story of the machines. Adoption then runs on a prisoner’s dilemma he takes from Nikhil Suresh’s July 2026 essay: no executive can report that AI did nothing without implying their peers lied about it. The 92% accuracy figure he cites comes from that essay, where Suresh reports it as Snowflake staff’s best case for their own database chatbot.

On valuations built from stories, two of his examples need adjusting. Meta’s one-day loss of about $230bn in February 2022, then a US record, followed its fourth-quarter results, not a first-quarter report (Variety). Procter & Gamble cut about $200m of digital advertising in 2017 without losing growth, which was a cut rather than the zeroing-out he describes (Adweek).

His way out is organised power. He sees rents against profits as a live split in the ruling class, visible in open against proprietary AI, and says labour should exploit it. His three levers are civil society (EFF, the Open Rights Group), occupying parties (Zack Polanski’s Greens as the model) and occupying unions. His union example is Shawn Fain’s UAW, whose Detroit Three contracts were set to expire on 30 April 2028 so other unions could align with them (WDET). The channel’s framing, that Altman, Amodei and Musk had agreed to slow down, goes further than the reporting: Time found on 15 September that no leading lab had yet truly slowed.