Leading AI Researcher Quits Over Concerns Industry Race Threatens Human Safety
San Francisco, Wednesday, 9 September 2026.
AI researcher Jacob Coxon resigned from Anthropic, warning that fierce corporate competition is forcing top labs to rush superintelligence development without solving existential safety risks.
Resignation Highlights Safety Discrepancies
Jacob Coxon, a 27-year-old researcher specializing in pretraining models, publicly announced his resignation from Anthropic on 8 September 2026 [1][4]. Coxon, who spent approximately three years working across both Anthropic and OpenAI, indicated his tenure began around 2023 following his transition between the two major laboratories [4][5]. In his announcement, he stated that neither company is acting responsibly regarding the development of self-improving superintelligence, describing the current industry trajectory as gambling with human lives [1][4]. His departure marks a significant moment for the sector, as Anthropic has previously positioned itself as a leader in AI safety governance compared to its competitors [3][4].
Internal Culture and Existential Risk
Internal communications reveal a stark contrast in how safety risks are perceived within the industry’s top firms. Coxon noted that while OpenAI staff may not have fully internalized the civilizational stakes, Anthropic employees understand the risks but feel locked into a race to deploy first [1][2]. Evan Hubinger, Alignment Science Lead at Anthropic, corroborated the severity of the situation, confirming a belief that AI poses a greater than 10% probability of causing human extinction within the next decade [2][3]. Hubinger admitted that despite these risks, the company does not yet have a concrete plan to solve alignment for superintelligence and is not clearly on track to do so [2][3].
Market Pressures and Future Trajectories
The competitive landscape is driving rapid deployment despite unresolved safety protocols, with researchers increasingly using terms like crunchtime and endgame to describe the pace of capability progress [3][4]. Coxon warned that without meaningful coordination across the industry or government intervention, situations could become out of control by the end of next year [3][4]. This tension exists as Anthropic prepares for a potential initial public offering and continues developing advanced models like Mythos 2, which industry figures suggest will be ready soon [1][3]. The consensus among concerned researchers is that attempting to speedrun alignment requires extraordinary confidence that there are no better trajectories available, a confidence currently lacking in private company Slack channels [1][4].