Live·Open questions in longevity research
All news
Industry & policyEcosystem

More than a hundred scientists, lawyers, and technologists sign “Anthropic Is Not a Safe AI Company” statement ahead of its IPO

19 November 2026· 261119004

More than a hundred scientists, lawyers, and technologists sign “Anthropic Is Not a Safe AI Company” statement ahead of its IPO

On November 18, the day before Anthropic shares began trading, the Center for AI Safety (CAIS) announced a statement. Its signatories argue that Anthropic, the company behind Claude, warns of the dangers of AI that builds successive AI systems itself (recursive self-improvement), yet seeks to be the first to build it. They say the company uses its reputation for safety to attract investors, employees, and public support, has weakened its safety commitments, and has lobbied for laws that serve its interests.

Signatories include psychologist Steven Pinker and Guillaume Verdon, founder of the e/acc movement for accelerating AI development. CAIS executive director Dan Hendrycks said: “People in this coalition disagree on almost everything, but we all see the danger in a company that has convinced itself that its dominance is the highest good.”

The argument about the race draws on Anthropic’s own documents. In September, its CEO, Dario Amodei, wrote about recursive self-improvement: “Left unchecked, it could outpace our ability to understand and control these systems, so it should be pursued very cautiously, if at all.” In its IPO filing, the company describes the cycle in which each version of Claude helps build the next as “one of its most significant structural advantages.” Max Tegmark, co-founder of the Future of Life Institute (FLI), an organization focused on AI risks, writes about Anthropic’s race toward self-improvement:

“They are two-faced: publicly, their leaders warn that it will produce uncontrollable systems and could lead to human extinction within ten years; behind closed doors, they are pushing AI self-improvement at full speed, accelerating the race, and boasting to prospective investors that this is their ‘competitive advantage.’”

As evidence of weakened commitments, CAIS cites the February update to Anthropic’s Responsible Scaling Policy, which sets out its safety rules. According to an analysis by the think tank GovAI, earlier versions included a commitment to pause development if the necessary safeguards were absent. The new version removes that commitment and promises to delay development only if Anthropic is clearly ahead or has yet to catch up with competitors that have strong safeguards. The company argues that if one developer stops while others continue without strong safeguards, the world could become less safe because the developers with the weakest safeguards would set the pace. FLI president Anthony Aguirre calls on Anthropic to stop this race immediately and secure legislation requiring its competitors to stop as well.

The statement identifies two dangers: concentration of power and loss of control, and signatories differ over which matters most. Tegmark emphasizes loss of control. Immunologist Derya Unutmaz wrote that he signed because of the concentration of power: “The real danger in the age of superintelligence is the concentration of this enormous power in one or two companies.” Thomas Wolf, co-founder of the AI platform Hugging Face, explained why he signed: Anthropic is on course to become the most powerful company on Earth. Such power requires a level of transparency equal to, or perhaps greater than, that expected of a sovereign government, and he fears the company is not yet moving in that direction.

Hendrycks added that many AI safety groups privately agreed with the statement but could not sign without jeopardizing relationships and future funding; Anthropic has “almost, but not entirely, captured the ecosystem meant to hold it accountable.” Anthropic did not immediately respond to a request for comment from The National.

Originally published on Telegram by Ukhvat NewsView on Telegram
Sources
#anthropic#ai-safety#recursive-self-improvement#responsible-scaling-policy#power-concentration#ipo