Anthropic's Claude Fable 5.1 (Opus 4.5-class model) launches on Amazon Bedrock and Claude Platform on AWS, alongside new "Enterprise Frontier Safeguards" data control option
Claude Fable 5.1 became generally available on Amazon Bedrock (via US and Global CRIS inference profiles, plus GovCloud) and on Claude Platform on AWS in North America. It is designated an Anthropic "Covered Model," requiring aws_review data retention mode (prompts/outputs retained up to 30 days for human safety review within AWS). Separately, AWS and Anthropic announced "Enterprise Frontier Safeguards," letting eligible customers use Fable 5 and 5.1 with zero data retention (ZDR) for internal use through Dec 31, 2026, with additional customer-controlled-key/audit features promised later in the year.
Entities: Anthropic, Amazon Web Services, Claude Fable 5.1, Amazon Bedrock, Claude Platform on AWS, Enterprise Frontier Safeguards
1 primary
What happened
Anthropic's Claude Fable 5.1 (an Opus 4.5-class model) is now generally available on Amazon Bedrock (US and Global inference profiles, plus GovCloud) and on Claude Platform on AWS in North America. As a designated "Covered Model," it requires the aws_review data retention mode, meaning prompts and outputs are retained up to 30 days for human safety review inside AWS. Separately, AWS and Anthropic announced "Enterprise Frontier Safeguards," letting eligible customers run Fable 5 and 5.1 with zero data retention for internal use through 31 December 2026, with customer-controlled keys and audit features promised later this year but not yet shipped.
Why it matters
This is a decision point for regulated enterprises and anyone with strict data-handling requirements: the default deployment path now involves 30-day human-reviewable retention, which some legal and compliance teams will not accept without the ZDR carve-out. Enterprises already on AWS need to actively check eligibility for Enterprise Frontier Safeguards rather than assume standard access is compliant. For developers and researchers, this is mainly an incremental capability and access update, not a new mode of using the model.
What is noise
The capability claims (better reasoning, agentic coding, multi-hour autonomous task execution) come entirely from Anthropic/AWS with no benchmark numbers, pricing, or independent verification, so they should be treated as marketing until third-party evals appear. This is also a same-day, expected cloud rollout for a point release, not a surprising or novel distribution event. The most consequential parts of Enterprise Frontier Safeguards (customer-controlled keys, full audit logging) are promised, not delivered, so treat them as roadmap items rather than current features.
Watch next
- 01Whether AWS/Anthropic publish concrete benchmark scores or independent third-party evaluations for Fable 5.1 versus Fable 5 and Opus 4.5
- 02Actual rollout and eligibility criteria for Enterprise Frontier Safeguards customer-controlled-key and audit-logging features before end of 2026
- 03Enterprise adoption signals, such as named large customers or case studies citing the ZDR option as a deciding factor for AWS/Anthropic over other vendors
Coverage
1 storyMore capability signals
Full feed →- AI systems outperform expert humans in persuasive communication22 Jun 202681
- Benchmark results show significant improvement in AI agent performance on WorkBench15 Jun 202679
- Introduction of Stateful ReAct Agents for Token-Efficient Autonomous Experimentation16 Jun 202678
- Study reveals flaws in LLM-as-judge safety evaluations due to temperature settings26 Jun 202677