Fable 5 is back online as of today. I already covered the lifting of the export controls yesterday — but the really interesting part is buried in Anthropic’s technical blog post about the redeployment. It goes way beyond just flipping the switch back on.
What Changed Under the Hood
During the 18-day suspension, Anthropic trained a new safety classifier. It blocks the specific jailbreak technique from the Amazon report in over 99% of cases. When a request gets blocked, users see a notification — and the request gets rerouted to Opus 4.8 instead.
The tradeoff: the new classifier triggers more often on harmless coding and debugging tasks. More false positives. Anthropic says they’re working on better calibration.
One interesting detail: Anthropic tested whether other models could find the same vulnerabilities as Fable 5 did in the Amazon report. The answer? Pretty much all of them could — from Haiku 4.5 to GPT-5.5 to Kimi K2.7. The reported technique didn’t actually unlock any unique capabilities.
The Jailbreak Severity Framework
This is the real news. Anthropic is proposing — together with Amazon, Microsoft, and Google — an industry-wide framework for consistently assessing jailbreak severity. Think of it as CVSS, but for AI model jailbreaks.
Four criteria would score every jailbreak:
Capability Gain — How far beyond existing tools does the jailbreak take you? Breadth — Does it work for one narrow task or across many? Ease of Weaponization — Does it need expert prompting and many retries, or does it work on the first try? Discoverability — Is the technique specialist knowledge or already publicly available?
This might sound like paperwork, but it’s actually a missing piece. Right now there’s no standard for how seriously a jailbreak should be taken. That creates uncertainty — for developers and governments alike.
Availability and Pricing
For Pro, Max, Team, and select Enterprise plans, Fable 5 is included for up to 50% of weekly usage limits through July 7. After that, it runs on usage credits. Access on AWS, Google Cloud, and Microsoft Foundry is being re-enabled as quickly as possible.
Anthropic also launched a new HackerOne program for security researchers to submit cyber jailbreaks they discover in Fable 5.
My Take
The blog post reads like a carefully crafted statement: ‘We take this seriously, but please don’t panic.’ The strategy of setting an extra-wide safety margin — blocking too much rather than too little — makes sense, but it will frustrate users. Whether the jailbreak framework actually gets adopted industry-wide remains to be seen. But having Amazon, Microsoft, and Google already at the table is a good sign.
Sources: Anthropic Blog: Redeploying Claude Fable 5 · HackerOne Program