AWS and Anthropic release Claude Fable 5 on Bedrock with stronger guardrails

AWS announced that Anthropic's Claude Fable 5 models will be available on Amazon Bedrock starting tomorrow, featuring significantly stronger guardrails for safe deployment. The release is part of Project Glasswing, a collaboration between AWS, Anthropic, and other industry partners to balance the powerful new cybersecurity capabilities of frontier models. The models offer advanced reasoning in most domains, but their guardrails are designed to prevent adversaries from gaining meaningful new vulnerability research capabilities. When a guardrail is triggered, the system automatically falls back to Claude Opus 4.8, which is already publicly accessible and itself a world-class model.

This cautious approach reflects AWS's commitment to both innovation and responsible AI deployment. The company emphasized that preventing adversaries from accessing deep vulnerability research tools is the highest priority. Anthropic has published a separate blog outlining their severity structure and SLAs for responding to reported issues with cyber-capable models. AWS's AI Red Team worked closely with Anthropic to further refine Fable’s protections, ensuring the model delivers strong reasoning while minimizing misuse risk. AWS views this as an iterative process, promising ongoing collaboration with partners to refine guardrails as new models and capabilities emerge.