US Gov Reviews AI Models Before Release: Anthropic Case

The U.S. government is now reviewing advanced AI models before their public release. This shift became clear after Anthropic released its Mythos 5 and Fable 5 models. The government suspended access to these models within days and subjected them to technical reviews. The immediate issue was a reported jailbreaking vulnerability. Here's the thing: The intervention started when Amazon CEO Andy Jassy alerted Treasury Secretary Scott Bessent to concerns. Anthropic then sent engineering teams to Washington to address the vulnerability. However, federal reviewers, including the NSA and the Center for AI Standards and Innovation, still required additional safeguards. Anthropic introduced new protections, which they say block the reported jailbreak technique 99% of the time. They also committed to collaborating with federal agencies on industry frameworks for evaluating AI systems. The models returned in stages. Mythos 5 first returned for about 100 U.S. companies and agencies, with Fable 5 following later after additional approvals. The Commerce Department formally removed the licensing requirement, but with a condition: the government can restore restrictions if needed. What's interesting is this episode has already influenced other AI developers. OpenAI adjusted its GPT-5.6 rollout, coordinating with federal officials for a limited preview. The bottom line: The government's role in vetting advanced AI is growing, impacting how these powerful technologies reach the public.