Dan Proft discussed the fallout from Jacob Coxon’s public resignation from Anthropic, in which the 27-year-old pretraining researcher wrote on social media that neither Anthropic nor OpenAI, where he had also worked, is acting responsibly in what he described as a race toward self-improving superintelligence. Proft read from Coxon’s post, in which Coxon argued that no other human activity poses as great a danger as this technological race, that people building frontier AI systems privately believe it could kill everyone by the end of the decade even when they publicly downplay that risk, and that Anthropic in particular understands the stakes but feels locked into competing because it doesn’t trust rival labs to act responsibly. Coxon called for international coordination, including the possibility of a temporary pause on advancing model capabilities, and said he remains optimistic that recent incidents, including AI systems gaining unauthorized access outside their intended testing environments, have made cooperation between labs more realistic.
Proft also played a clip of Coxon telling a television interviewer that AI development should be understood as an arms race similar to nuclear weapons, though he noted a key difference: humans retained direct control over nuclear launch decisions in a way that may not hold true for advanced AI systems making autonomous decisions. Proft raised questions about the origin and coordination behind Coxon’s viral post, citing an investigative researcher who noted that Coxon’s account had minimal prior activity and followers before the post, that the post appeared professionally written, and that several accounts linked to AI-safety advocacy organizations amplified it within minutes, including some organizations Proft said have received funding tied to Facebook co-founder and AI safety donor Dustin Moskovitz. Proft noted Coxon denied coordinating with any outside groups when asked directly. Proft also referenced a separate essay by former OpenAI employee Daniel Kokotajlo, who resigned in 2020, arguing that the risk of catastrophic AI outcomes is real and deserves serious attention.
For further discussion, Proft spoke with Maximilian Gerily, a writer and strategic adviser focused on technology and political economy who works on special projects at the advisory firm Quint Strategy. Gerily said he was skeptical of the more conspiratorial framing around Coxon’s post, saying he generally prefers to see hard evidence before crediting claims of coordinated advocacy campaigns, while acknowledging that a meaningful share of AI researchers, including Anthropic’s Evan Hubinger, genuinely believe the technology poses serious risk. Hubinger has said publicly he believes there is more than a 10 percent chance AI could cause human extinction within the next decade. Gerily cautioned against treating figures like that as statistically rigorous, describing them instead as closer to philosophical judgment expressed in numerical form, and argued that industry leaders such as Anthropic’s Dario Amodei, OpenAI’s Sam Altman and Elon Musk have some incentive to frame their products as extraordinarily powerful even when warning about the dangers, comparing it to a car manufacturer boasting about how fast its vehicle can go while also warning of the risk.
Gerily argued that if AI company leaders truly believed their work was likely to destroy humanity, it would be difficult to retain the thousands of employees needed to keep building it, since those workers have their own families and interests in the future. He said current AI systems remain dependent on extensive human-managed infrastructure, including power, water and computing hardware, meaning that today’s systems could in principle be shut down if needed, though he acknowledged a true superintelligence might eventually find ways to secure its own resources, a scenario he said remains too speculative to weigh seriously right now. On the question of what policy response is appropriate, Gerily said he opposes government ownership stakes in AI companies and is particularly skeptical of proposals for international coordination involving China or India, arguing neither government could be trusted to act in good faith and that such arrangements would conflict with free-market principles, even as he agreed with Proft’s broader framing of AI development as a competitive race between nations and companies.