Anthropic and the White House are working on a framework to assess the severity of AI security flaws and guide when the government should step in, according to POLITICO, a shift from the standoff that pulled Anthropic's newest models offline. The talks follow export controls the White House imposed over a so-called jailbreak in Fable 5 and Mythos 5, a flaw the two sides disagreed on, which forced Anthropic to suspend access to both models for all users. Negotiators aim to set common benchmarks for judging future incidents, including how far safeguards were bypassed, what capabilities were exposed, and the real-world consequences of a breach. Talks had collapsed Friday after Anthropic refused to de-deploy Fable, but a weekend of calls and a week of meetings in Washington revived them. The export controls have not yet been lifted.
