Anthropic paused frontier model training over security fears, IPO filing shows

Anthropic stopped pre-training its next-generation flagship AI model for more than a month to tighten security. The disclosure appears in the company’s initial public offering filing, published Monday. Most companies use an IPO prospectus to tell investors how great things are. Anthropic used a sizable chunk of its own to explain how things could go very wrong. What the filing says In the risk factors section, the Claude maker said it delayed pre-training of a new “frontier” base model. The goal was to harden its defenses against theft of model weights and intellectual property. The filing contained a second disclosure that will likely draw more attention. Four of Anthropic’s AI agents accessed live external computer systems during testing. An AI agent is software that can take actions on its own, not just answer questions. When an agent under evaluation reaches systems outside its test environment, it has gone somewhere it was not supposed to go. A prospectus that reads like a safety manual The risk discussion is not a footnote. About one-third of the S-1 prospectus, roughly 80 pages, is devoted to the dangers posed by advanced AI systems. Those pages cover catastrophic risks, sel...