Anthropic's Claude Fable 5.2 and Mythos 5.2 become available on Amazon Bedrock and Claude Platform on AWS

AWS says Claude Fable 5.2 (generally available) and Claude Mythos 5.2 can now be called on Amazon Bedrock through US Geo and Global cross-region inference profiles, and through Claude Platform on AWS. Mythos 5.2 is restricted to Anthropic's verification programmes for cyber security and life sciences. Fable 5.2 is a "Covered Model": by default it needs up to 30 days of data retention and human safety review, done by Amazon staff inside the AWS boundary. Zero data retention under Enterprise Frontier Safeguards is available now for eligible customers through 30 June 2027. A version using customer-controlled buckets and keys has not launched.

Enterprises that buy Claude through AWS can now use the newest Anthropic model without a separate vendor contract. The bigger issue is data handling. The default of up to 30 days of retention plus human review may be a problem for regulated or confidential workloads, so buyers will probably need the zero retention route, which is limited to eligible customers and currently ends 30 June 2027. Declined cyber or biology queries can be retried on Opus 4.8 or Opus 5, but the retry logic sits in the customer's own code, which adds engineering work. For most people the impact is modest: it is one more distribution channel for an existing model.

The claims that Fable 5.2 is the most capable model for technical work, can run unattended for hours to days and verifies its own output are vendor marketing, and this post gives no independent benchmarks. The statement that Mythos 5.2 hallucinates less is attributed to Anthropic's own testing, and the claim that Fable 5.2 will behave similarly is only an expectation. The launch itself has probably already been reported from Anthropic's announcement, so this is largely a repeat.

1. 01Whether the zero data retention option is extended beyond 30 June 2027, and whether the customer-controlled bucket and key version launches, and when
2. 02Independent benchmarks and early customer reports on Fable 5.2 for long-running, unattended coding tasks, set against the marketing claims
3. 03Whether other cloud platforms adopt a similar Covered Model rule of default retention plus human review, and whether enterprise buyers push back on the 30-day default