Anthropic's Fable 5 Is Back — After Amazon Got It Banned

By: Anton Kratiuk | today, 13:51

Anthropic's Fable 5 AI model is available again globally as of July 1, after the US Commerce Department lifted the export restrictions it imposed on June 12. The ban lasted 18 days and cut off all users worldwide — not just non-Americans, as the government had originally requested, because filtering by citizenship was technically impossible to implement. The return comes with a permanent safety catch built in.

How the ban started

Amazon researchers discovered a jailbreak technique: a way to prompt Fable 5 into finding software vulnerabilities and writing working exploit code. Amazon CEO Andy Jassy personally called Treasury Secretary Scott Bessent on June 12 to flag it as a national security concern, per TechTimes. The Commerce Department moved the same day, ordering Anthropic to restrict access.

The conflict of interest here is hard to miss. Amazon has invested $13 billion in Anthropic, hosts it on AWS, and simultaneously competes with it through its own Nova model line. A company reporting its own major investment to federal regulators — and triggering a shutdown that hurt that investment — is an unusual dynamic, to say the least.

What changes now

Anthropic spent the 18 days building a new safety classifier. The company says it blocks the jailbreak method described in Amazon's report in more than 99% of cases. When a blocked request is detected, it gets rerouted to Opus 4.8, and the user is told why. Anthropic confirmed the rollout plan includes a condition set by Commerce Secretary Howard Lutnick: Anthropic must report malicious activity and coordinate future model releases with the government going forward — a permanent oversight arrangement, not a one-time fix.

The return is phased. Paid plan users face 50% usage limits through July 7, after which full access switches to a credits system. Access on AWS, Google Cloud, and Microsoft Azure is coming back too, but Anthropic hasn't given firm dates for any of those platforms.

The bigger picture

This is the first time the US government has used export controls to effectively kill-switch a commercial AI model. The precedent matters: OpenAI's GPT-5.6 is reportedly facing a similar staged rollout review. Commerce Secretary Lutnick's language — that Anthropic "worked closely" with the government — signals that front-line AI labs now need federal sign-off before major releases, without any formal law requiring it. That's a new kind of regulatory power, built on ad-hoc authority rather than legislation.