The Cipher

Late Friday afternoon US time, Anthropic received a government directive it couldn't negotiate around: suspend all access to Fable 5 and Mythos 5 for foreign nationals — inside the US or out — effective immediately. By evening, both models were dark for every user on earth, because there was no clean technical way to enforce the restriction without killing them entirely.

The stated reason was national security. The Commerce Department cited a discovered jailbreak technique that it believed could allow adversaries to extract dangerous capabilities from the models. Anthropic pushed back hard and publicly, writing that the jailbreak was "narrow" and that the capability it exposed was "widely available elsewhere." The company said it disagreed the finding should trigger a commercial recall affecting hundreds of millions of users, but it complied anyway.

Here's where it gets more complicated. TechCrunch reports that Amazon CEO Andy Jassy may have been the one who raised the original flag. Amazon is both a major investor in Anthropic and the company's primary cloud partner — AWS is how most of the world accesses Claude. If that reporting holds, a commercial partner effectively triggered a government shutdown of a competitor's most valuable product. That's a dynamic worth watching very carefully.

The broader precedent is what should keep AI lab executives up at night. The US government has now demonstrated it will use export control machinery — the same framework it uses to restrict semiconductor sales to China — to pull an AI model from the market at short notice, with limited explanation, and with global effect. Anthropic didn't get advance warning. It didn't get a detailed threat briefing. It got a 5:21 p.m. Friday order.

For Australian users and researchers, both models were simply unavailable from Saturday morning. Australia doesn't sit inside the "foreign national" carve-out — there is none. Australian companies and universities using Fable 5 or Mythos 5 via the API woke up to broken integrations with no timeline for restoration.

The irony here is pointed. Anthropic has built its entire brand identity around being the safety-conscious AI lab — the one that publishes responsible scaling policies, that runs red-team exercises, that flags its own models' risks. Now its most public safety disclosures appear to have been handed to regulators as ammunition. Being transparent about what your model can do, it turns out, also tells governments what they might want to suppress.

What to watch: Whether Fable 5 and Mythos 5 come back in a restricted form, whether other frontier labs (OpenAI, Google DeepMind) face similar orders for their most capable models, and whether the Commerce Department publishes any detail on the actual threat that justified this. Also watch for how Australia's AI safety discussions respond — the government has been developing its own AI governance framework, and this event will feature prominently in those conversations.