Anthropic launches Enterprise Frontier Safeguards for secure AI monitoring

Anthropic introduced a 30-day data retention policy starting with Fable 5 to better detect sophisticated misuse that spans multiple sessions and accounts. This policy, however, created adoption barriers for regulated industries, which often require strict access controls on sensitive data. Enterprise Frontier Safeguards is a compromise: it enables continuous misuse detection without giving Anthropic direct access to the data, since monitoring data stays in the customer's cloud, and flags are sent directly to the customer's team.

Designed with input from over 100 customers, including Bank of America and other large banks through the ARC industry group, as well as Comcast, KPMG, Mastercard, Salesforce, and Visa, EFS reflects a growing demand for AI tools that meet enterprise security and compliance requirements. While customers pay for storage and data transfer, the feature itself is free, and it is available across major platforms like Amazon Bedrock and Google's Agent Platform. This launch, alongside the new Claude Fable 5.1 and Claude Mythos 5.1 models, signals Anthropic's push to serve enterprise clients with both advanced AI capabilities and robust governance controls.