Anthropic's brand new AI models are gone. Both Mythos 5 and Fable 5 — released just days earlier with significant fanfare — were completely shut down Friday night after the Trump administration issued an emergency directive. The company had no choice but to flip the kill switch for all customers, a dramatic move that signals how seriously the government is now treating AI security threats.
Here's what happened: The Commerce Department sent Anthropic a directive Friday evening imposing strict export controls on the new models, effectively confining them to US use only. But there's a catch — Anthropic couldn't technically comply with those restrictions without completely disabling the models for everyone. So they did. In a terse Friday night announcement, the company explained that shutting down Mythos 5 and Fable 5 entirely was "the only way" to ensure immediate compliance with the government order. All other Anthropic models remain available.
The trigger? A reported security vulnerability. According to an administration official speaking to Axios, the government got wind of a jailbreak — a workaround that lets users bypass Fable 5's safety guardrails — specifically to extract dangerous information about cybersecurity, chemistry, and biology. The alleged exploit works by tricking the model into reviewing a specific codebase for vulnerabilities. The administration essentially asked Anthropic to pump the brakes so the "national security apparatus" could be hardened against this kind of threat. A source suggested that hardening could take "a few weeks."
Now here's where it gets interesting. Anthropic pushes back on the severity. The company says the government only showed them "verbal evidence" of a "narrow, non-universal jailbreak" — meaning it's not a widespread exploit that works consistently. More importantly, Anthropic claims the jailbreak has only been used to find minor and relatively simple software bugs, nothing catastrophic. They also point out that OpenAI's GPT-5.5 can do basically the same thing without triggering government intervention. So why the hammer on Fable 5?
This moment matters because it reveals how the new administration plans to regulate AI — not through legislation or careful policy, but through sudden, forceful directives that can shut down entire products overnight. For companies like Anthropic, it's a sobering reminder that moving fast and breaking things doesn't work when the government decides your creation is a national security risk. For the rest of us, it raises a bigger question: if a potential jailbreak on a newly launched model warrants this kind of response, what does AI regulation actually look like going forward? The answer might unfold over the next few weeks as Anthropic works to bring these models back online — if they ever do.