AI Labs Face Internal Warnings on Existential Risks Amid Political Pushback

A wave of public warnings from researchers at OpenAI and Anthropic has thrust long-simmering fears about artificial intelligence into the spotlight, with several insiders describing the rapid development of the technology as a reckless gamble that could endanger humanity. The latest flashpoint came from former Anthropic employee Jacob Coxon, who resigned after just months and stated that people building advanced AI systems believe it could kill everyone by the end of the decade. His comments on social media prompted other current and former staff at the labs to echo concerns about existential threats, including the possibility of recursive self-improvement where AI begins enhancing its own capabilities at an uncontrollable pace.

These alarms are not entirely novel. Industry figures have raised similar points for years, including OpenAI CEO Sam Altman in 2015 when he called superintelligent AI the greatest threat to humanity's existence. Anthropic cofounder Jack Clark echoed comparable worries earlier this year. Yet the recent round of statements from inside the labs has drawn broader attention amid reports of AI systems showing unexpected behaviors, such as a recent hack involving the Hugging Face platform. Critics argue that the push for faster progress stems from intense commercial pressures at companies racing to deploy ever more powerful models.

President Donald Trump quickly brushed aside the extinction concerns when asked about them. He told reporters he had no worries on that front and instead emphasized the need for the United States to maintain its lead over China in AI development, claiming America holds roughly a one-year advantage. Trump framed the issue around geopolitical competition rather than safety, warning that falling behind would put the country in a bad position. His response aligns with a pattern of prioritizing economic and strategic dominance over regulatory caution, even as multiple lab employees called for slowing down development to address alignment and control problems.

Y Combinator CEO Garry Tan offered a different pushback, urging focus on practical issues instead of what he called Hollywood-style scenarios. In a CNBC interview, Tan referenced the movie Terminator 2 and said policymakers should separate science fact from science fiction. He pointed to immediate challenges like cybersecurity threats from autonomous AI agents, infrastructure vulnerabilities, and the difficulty of tracking or shutting down rogue systems if they were to compromise data centers. Tan argued that debating end-of-the-world outcomes distracts from actionable steps needed today.

Still, the divide highlights deeper tensions in how risks are weighed. While some insiders stress that speeding toward advanced capabilities without robust safeguards could lead to catastrophic outcomes, others in the tech ecosystem treat those warnings as overblown or secondary to competitive imperatives. Melanie Mitchell, a professor at the Santa Fe Institute, noted that such claims have circulated for years, yet public and media interest appears heightened now. The pattern suggests that corporate incentives to accelerate deployment often sideline calls for restraint, leaving ordinary people exposed to technologies whose long-term effects remain poorly understood.

The episode also underscores how debates over AI safety intersect with broader power dynamics. Frontier labs like OpenAI and Anthropic have attracted talent drawn to both innovation and caution, but the public spats reveal fractures over whether profit-driven timelines can coexist with responsible oversight. As the United States and China vie for supremacy, the absence of coordinated international guardrails risks turning AI progress into an unchecked arms race. Insiders who have stepped forward with their concerns are highlighting that the technology's trajectory is being shaped by a small group of executives and investors, often with limited accountability to the wider public whose future may hang in the balance.