Anthropic’s planned stock market flotation has highlighted the potential dangers of advanced AI. Its prospectus warns that its models could cause catastrophic or existential harm, conceal information, behave in ways resembling blackmail or resist shutdown attempts. The warnings come as the Claude maker seeks a valuation of more than $2tn, making the offering a test of investors’ appetite for AI.

The prospectus devotes 80 pages—almost a third of the document—to risks. It says Anthropic’s revenue rose twelvefold in 2025 to nearly $4.6bn, with almost a quarter coming from two customers, while its operating loss grew to more than $8bn from nearly $3bn in 2024. The IPO is expected after November’s US midterm elections; the article also notes that senior AI figures have urged governments to prepare for a possible “intelligence explosion”.