Anthropic introduces 30-day data retention policy for Claude Fable 5 model

09:39

By: Dakir Madiha

Anthropic introduces 30-day data retention policy for Claude Fable 5 model

Anthropic has released Claude Fable 5, its most advanced AI model aimed at public and enterprise users, alongside a major shift in data governance rules. The company now requires a 30-day retention period for all inputs and outputs processed through the Fable 5 system, marking a significant change from previous zero-retention agreements offered to enterprise clients.

The new policy applies across multiple access points, including proprietary interfaces such as claude.ai and Claude Code, as well as third-party cloud platforms like AWS Bedrock, Google Cloud, and Microsoft Azure. Organizations that previously negotiated zero-data-retention contracts must now accept the updated terms to use Fable 5. On AWS, access is explicitly conditional on compliance with the retention requirement, effectively making it a gatekeeping rule for deployment.

Anthropic states that retained data will not be used for model training. The company frames the 30-day storage window as a security measure designed to counter emerging threats. These include attempts to bypass safety systems, model distillation attacks, and state-linked exploitation efforts. The company also allows limited human review of flagged conversations under strict access controls, with activity logging and restrictions on data export.

The policy includes automatic deletion after 30 days unless legal obligations or active security investigations require extended retention. Some enterprise users can apply additional safeguards such as customer-managed encryption keys and access audit logs. Despite these controls, the shift signals a clear departure from earlier privacy commitments that emphasized zero retention for enterprise workloads.

Industry observers note that the decision could influence broader AI governance standards. As frontier models gain more capability, providers are increasingly linking access to stricter data handling rules framed as safety requirements. For enterprises integrating advanced AI into sensitive workflows, the trade-off now includes accepting temporary data storage in exchange for access to higher-performance systems.