For three weeks, the most capable AI model available to the public simply didn't exist.

Anthropic released Claude Fable 5 on June 9. By June 12, the US government had slapped export controls on it, and because Anthropic had no way to verify anyone's nationality in real time, the company pulled the model for everyone. Not just foreign users. Everyone.

As of July 1, it's back. But the version of events Anthropic published alongside the redeployment tells a story that matters far beyond one model, because it's the first time a government has effectively switched off a commercial AI product, and the first real preview of how AI releases might work from here on out.

Amazon found the crack

The shutdown traces back to an unlikely source: Anthropic's own partner.

Amazon researchers discovered a way to prompt Fable 5 past its safety guardrails, getting it to identify software vulnerabilities and, in one case, produce code showing how a vulnerability could be exploited. That report reached the US government, and the export control order followed almost immediately.

Here's the part Anthropic is keen to stress. When they tested the same technique against other models, including their own older ones, OpenAI's GPT-5.5, and Kimi K2.7, every single one could produce the same exploit demonstration. The jailbreak didn't unlock anything unique to Fable 5. It poked through the outer layer of a deliberately oversized safety buffer, into territory Anthropic describes as routine defensive security work.

In other words, the model got benched for something its competitors could already do. That's the awkward reality of being the frontrunner. The scrutiny lands on you first.

What Fable 5 actually is

Some quick context helps here. Fable 5 and its sibling, Mythos 5, are the same underlying model wearing different amounts of armor.

Mythos 5 ships with fewer restrictions and goes only to a small group of vetted organizations doing defensive cybersecurity work, under a US government program called Project Glasswing. Anthropic openly says Mythos can find and exploit software vulnerabilities better than any other model and all but the most skilled human experts.

Fable 5 is the public version, wrapped in the heaviest safeguards Anthropic has ever shipped. The company doubled its safeguards team in the month before launch and deliberately tuned its safety filters to over-block. Their logic, better to frustrate users with false refusals than let one genuinely dangerous request through. They call it a safety margin. Users call it the model refusing perfectly normal requests, which Anthropic admits will now happen even more often, since the fix for the Amazon jailbreak involves a new classifier that flags benign coding and debugging tasks more frequently.

When Fable 5 blocks a request, it now routes it to Opus 4.8 instead. So you'll still get an answer, just from a less capable model.

What limited access means for you

The redeployment comes with strings attached, and they're worth knowing before you plan your workflows around it.

Fable 5 is available globally on the Claude Platform, Claude.ai, Claude Code, and Claude Cowork. For Pro, Max, Team, and select Enterprise plans, it's included for up to 50% of weekly usage limits, but only through July 7. After that, access runs on usage credits. Standard Enterprise seats don't get any included allowance at all. Cloud access through AWS, Google Cloud, and Microsoft Foundry is still being restored.

So this is less a full return and more a supervised release. You get a week of generous access, then the meter starts running.

The bigger shift nobody should ignore

The most consequential part of the announcement has nothing to do with pricing.

Anthropic is now working with Amazon, Microsoft, Google, and its Glasswing partners on a shared framework for scoring the severity of AI jailbreaks, similar to how the security industry scores software vulnerabilities. The idea is to give both companies and governments an agreed standard for when a jailbreak is a minor curiosity and when it's an emergency.

Alongside that, Anthropic committed to giving government partners early access to frontier models before release, sharing jailbreak intelligence quickly, and staffing dedicated teams for joint security research. The company is openly calling for these arrangements to be codified into regulation that applies to every frontier lab.

Read between the lines and the message is clear. The era of AI labs shipping their most powerful models on their own schedule is ending. Fable 5's three-week disappearance wasn't a glitch in the system. It was a preview of the system.