AWS makes Claude Fable 5.2 and Mythos 5.2 available on Amazon Bedrock

Amazon Web Services reports that Anthropic's Claude Fable 5.2 and Claude Mythos 5.2 are now offered on Amazon Bedrock and Claude Platform on AWS. Anthropic presents Fable 5.2 as its most capable generally available model for technical work, while access to Mythos 5.2 stays restricted to vetted programmes.
What the two models are for
According to the announcement, Claude Fable 5.2 is Anthropic's most capable generally available model for technical work, with improvements in coding and the sciences. It is described as suited to ambiguous, multi-step tasks, able to check its own output and capable of running unattended for hours to days.
Claude Mythos 5.2 is available only through Anthropic's verification programmes, for advanced work in cyber security and life sciences. The announcement names security engineering, red teaming and penetration testing, together with biomolecular design and research synthesis, as its focus areas.
Testing results and release timing
In Anthropic's alignment testing, Mythos 5.2 is reported to hallucinate less, complete tasks more reliably, surface important information more consistently and self-report issues more honestly than Opus 5.5, Sonnet 5.5 and Mythos 5.1. Anthropic released Claude Fable 5.2 on November 9.
Declined requests and fallback behaviour
When a cyber security or biology query is declined, the announcement says it can be sent again to Opus 4.8 or Opus 5 respectively. On Amazon Bedrock, however, the retry is left to the customer's code rather than happening as an automatic switch. If the declined request reached the prompt cache, the refusal also carries a fallback credit token, which can be sent with the retry and is billed as cache reads.
The system card, as cited in the announcement, flags a weakness: when a safety classifier passes a request to the older Opus 4.8, prompt injection attacks succeed more often. The announcement directs this at customers running Claude as an agent in Cowork, Claude Code or the browser.
Data retention and access
Anthropic classes Fable 5.2 as a Covered Model. Customers must set the aws_review data retention mode on their account or project before invoking it. Enterprise Frontier Safeguards, built in collaboration with AWS, combine zero data retention with the latest misuse-detection safeguards. A version that keeps data in a customer's own bucket under the customer's keys is expected but has not launched; until then, eligible customers can use Fable 5.2 with zero data retention through June 30, 2027.
Open questions
- When will the version that keeps data in customer-controlled buckets launch?
- How much more often do prompt injection attacks succeed when requests fall back to Opus 4.8?