Bryan Cantrill: AI doom fear spreads without evidence
A 27-year-old former Anthropic staffer puts humanity's extinction odds above 10 percent. Bryan Cantrill wants evidence, not contagious panic.
In short
Bryan Cantrill rejects the claim that AI could wipe out humanity by the end of the decade, because the mechanisms named are neither spelled out nor inside the expertise of the people raising the alarm.
At a glance
- Jacob Coxon, 27, formerly at Anthropic: more than 10 percent odds that AI kills every human in the next decade.
- Evan Hubinger agrees with that estimate, according to Cantrill's post.
- Cantrill published on September 13, 2026; Simon Willison linked to it on September 14, 2026.
- Cantrill faults the phrases “hacking critical infrastructure” and “extinction-level bioweapons” as unelaborated.
- His yardstick is Carl Sagan: extraordinary claims require extraordinary evidence.
Bryan Cantrill rejects the claim that AI could wipe out humanity by the end of the decade: the mechanisms offered in support are neither spelled out nor inside the expertise of the people putting them into circulation.
What set it off
Jacob Coxon, 27, who used to work at Anthropic, puts the odds that AI kills every human in the next decade at more than 10 percent. Evan Hubinger agreed with that estimate. Cantrill posted his reply on September 13, 2026, and Simon Willison carried it in his link blog on September 14, 2026.
The prank he still regrets
Cantrill opens by indicting himself. Late one night in his first year at university, in the week before finals, he and other computer science students convinced humanities classmates that a computer virus was tearing through the campus. The panic outran every attempt to walk it back, and coursework was lost. His shorthand for the dynamic comes from Jonathan Swift: falsehood flies while truth limps behind.
Expertise is borrowed credit
The disagreement is not really about a percentage. Being treated as an expert means drawing on trust the audience cannot independently audit, and Cantrill's position is that this credit must not be spent on claims outside one's own field. He invokes Carl Sagan's standard — extraordinary claims require extraordinary evidence — and finds none supplied here.
The objection from building hardware
Cantrill builds computers, and his counterargument is a physical one. Phrases such as “hacking critical infrastructure” or “extinction-level bioweapons” name a feeling rather than a route, he argues. Agency runs through machines that people design, power and control, and being smart does not suspend those constraints; he points readers to his 2023 Monktoberfest talk for the longer version.
What stays unconfirmed
Both sources read for this article report Coxon's statement, but the original tweet was not checked directly, so the exact wording comes second-hand from Cantrill and Willison — as does Hubinger's assent. Neither post establishes how widely the view is actually held inside Anthropic.
FAQ
Who is Jacob Coxon?
A 27-year-old former Anthropic employee whose statement about AI extinction risk prompted Cantrill's post.
What probability is being claimed?
More than 10 percent within the next decade, an estimate Evan Hubinger agreed with according to Cantrill.
What exactly is Bryan Cantrill objecting to?
Using phrases like “hacking critical infrastructure” and “extinction-level bioweapons” with no mechanism given and outside the speaker's own field.
