On June 30 the Commerce Department lifted the June 12 export-control directive on Fable 5 and Mythos 5, ending a 19-day shutdown of the most capable model ever pulled from the market.
Anthropic's redeployment post details the price of freedom:
- A new safety classifier blocking the Amazon-flagged vulnerability-discovery jailbreak in >99% of attempts
- A HackerOne bounty program for Fable jailbreaks
- Pre-release government review of future frontier models
- A cross-lab jailbreak-severity rubric built with Amazon, Microsoft, and Google
Fable 5 returned globally July 1; Mythos 5 came back only for select US organizations.
Axios published the behind-the-scenes of the standoff — including the detail, awkward for everyone involved, that Amazon CEO Andy Jassy reported the jailbreak to the Treasury Secretary before telling Anthropic. Amazon remains Anthropic's largest investor. Thanksgiving dinner will be tense.
What all the fuss is about
As a reminder of the capability that spent 19 days off the market:
- 80.3% SWE-Bench Pro
- 88.0% Terminal-Bench 2.1
- 59.0% Humanity's Last Exam (64.5% with tools)
- Cyber adversarial-robustness eval: attack success rate drops from Opus 4.8's 56.6% to 5.4%
Why it matters
The arc is unmistakable: the US government is no longer reacting to frontier AI releases — it's in the release pipeline. Between the classifier, the bounty program, and standing pre-release review, this isn't a one-off intervention; it's the template for how every frontier launch happens from now on.
Sources:
- Anthropic redeployment post
- Axios behind-the-scenes report on the standoff
About the Authors
Federico Ulfo
Founder, Engineer
New York City