White House & Anthropic Negotiate AI Security Rules

The White House is now negotiating with Anthropic to create a framework for evaluating security flaws in advanced AI models. This shift follows a Commerce Department directive that forced Anthropic to suspend global access to its Claude Fable 5 and Claude Mythos 5 models. The directive came after the government learned of a potential jailbreak of Fable 5. Administration officials and Anthropic CEO Dario Amodei disagreed on the severity of this jailbreak. Anthropic described the method as narrow and widely reproducible by other models. Anthropic's negotiation team includes Sarah Heck and Tom Brown. The goal is to establish common benchmarks for classifying AI security incidents and guiding government responses. This matters because it could shape how AI security issues are managed in the future.