A tweet by former Anthropic employee Jacob Coxon confirming that many Anthropic researchers believe AI could kill all humans by the end of the decade has spread into mainstream conversation. Coxon cited hacking critical infrastructure and extinction-level bioweapons as mechanisms. Bryan Cantrill, writing via Simon Willison's Weblog, calls these claims reckless.

Cantrill's core objection is epistemic. Coxon is not an expert on critical infrastructure, bioweapons, or extinction biology. Domain experts hold implicit public trust, and that trust is being spent on claims that rely on extrapolation rather than evidence. The burden of proof, Cantrill argues, sits with whoever is raising the alarm.

Watch whether Anthropic responds formally or lets the employee-sourced narrative calcify. Uncontested fear claims from credible-sounding insiders shape regulation faster than technical rebuttals. Cantrill's Oxide and Friends episode addresses the bioweapons concern in more depth starting at 51m44s.