Jacob Coxon, a 27-year-old pretraining researcher at Anthropic, has publicly resigned from his post, citing grave concerns regarding the accelerating pace of advanced frontier model development. Coxon, who previously conducted research at OpenAI, underscored his stance by voluntarily forfeiting unvested equity shortly before it was scheduled to mature. In his public statement, he accused leading artificial intelligence laboratories of engaging in an irresponsible contest, claiming that current commercial practices amount to gambling with human survival.
The focal point of Coxon's criticism is the industry-wide race toward recursively self-improving superintelligence, conducted in the absence of validated control mechanisms. According to Coxon, competitive pressure compels leading labs to rapidly train and deploy increasingly capable models while foundational alignment questions remain unanswered. His departure directly from the technical engine room of pretraining adds considerable weight to long-standing warnings about unchecked technological escalation.
Coxon received notable backing from within Anthropic itself shortly after announcing his departure. Evan Hubinger, the company's Alignment Science Lead, publicly agreed with the core premises of Coxon's statement on the social platform X. Hubinger affirmed that the theoretical and practical foundations required to reliably guide autonomous systems remain fundamentally incomplete. He noted that Coxon's frustration with prevailing development timelines is shared by researchers working directly on technical safety.
Hubinger went further by providing a stark quantitative evaluation of the danger facing humanity. He estimated the probability of artificial intelligence causing human extinction within the coming decade to be greater than ten percent, pointing directly to the lack of an existing alignment blueprint for superintelligence. This public admission from a senior safety scientist highlights that catastrophic risk scenarios are treated as imminent technical hurdles rather than distant philosophical thought experiments.
The resignation places renewed scrutiny on Anthropic's founding identity as a safety-first public benefit corporation. The lab was established by former OpenAI employees specifically to prioritize thorough alignment over commercial speed. Coxon's departure and Hubinger's candid remarks suggest that intense market competition is testing these initial ideals. As internal dissent becomes increasingly public, pressure is mounting on regulators to implement binding oversight rather than relying on voluntary industry assurances.