AWS has announced that Anthropic’s Claude Fable 5 models will be available on Bedrock starting tomorrow. These models include stronger guardrails to prevent misuse, ensuring customers receive the latest capabilities quickly while maintaining enterprise-grade security and privacy. Bedrock, which provides customers with world-class performance and the broadest selection of models available anywhere, continues to support AWS’s commitment to secure AI services. According to AWS, the release aligns with customer demand for immediate access to new models, combined with the trusted enterprise features of AWS. The announcement comes as part of AWS’s ongoing efforts to balance innovation with security, ensuring that frontier models are released responsibly and with the necessary protections in place. The latest generation of models, such as Anthropic’s Claude Mythos, has demonstrated powerful new capabilities, particularly in cybersecurity, as seen through Project Glasswing. This initiative involves close collaboration with Anthropic and other industry partners to refine guardrails for new models, ensuring they are both effective and secure. The goal is to prevent adversaries from gaining access to advanced vulnerability research capabilities while allowing companies, governments, and academic institutions to protect their assets first. AWS emphasized the importance of continuous improvement in guardrails as new models are released and as the industry learns more about their effectiveness. The company is committed to iterating with partners, delivering value, and responding to changes in the industry. It is also crucial to address any issues with these models after release, as highlighted by Anthropic’s recent blog on Redeploying Fable 5, which outlines their approach to issue severity and response for cyber-capable models. AWS’s AI Red Team has worked with Anthropic to further improve Fable’s protections, and they believe the latest guardrails result in a model that minimizes the risk of misuse by adversaries. When guardrails are triggered, the model automatically falls back to Opus 4.8, a publicly accessible model already in use. AWS appreciates Anthropic’s partnership and commitment to defenders, and looks forward to continued collaboration with the industry to ensure the safe and secure release of frontier models.
Source: awsml
