Updated Oct. 26, 2026, 3:41 a.m. ET: Adds details of Enterprise Frontier Safeguards, which Anthropic announced on Sept. 1.
Anthropic is changing how it handles data retention for its most capable models after pushback from enterprise customers. Bloomberg reported on Aug. 20 that the company planned to change its data retention policy for advanced AI, and Reuters, citing a source, reported the same day that Anthropic planned to change its enterprise data retention policy.
The dispute centers on a policy Anthropic introduced with Claude Fable 5: retaining enterprise prompts and outputs for 30 days so it could detect misuse. Anthropic has said the policy “was not motivated by a desire to train on enterprise data,” and that it “has never trained on enterprise data without explicit permission, and never will.”
Why Anthropic Wanted the Data
In a post announcing the new approach, Anthropic said Mythos-class models bring “the potential for both misuse and autonomous misbehavior,” and that it had seen substantial evidence of attempted misuse, ranging from fraud to sophisticated cyberattacks, including cases involving stolen enterprise credentials. Because the most sophisticated misuse can be spread across many sessions and accounts, the company argued, analyzing each interaction and immediately discarding it is not enough. Detection requires storing data long enough to correlate activity over time.
Enterprises generally understood the security value, Anthropic said, but many, “especially in regulated industries,” found it difficult to use models with data retention.
Enterprise Frontier Safeguards
Anthropic’s answer is Enterprise Frontier Safeguards (EFS), developed with more than 100 customers in financial services, healthcare, manufacturing, telecom, law, retail and the public sector, along with AWS, Google Cloud and Microsoft Azure. Companies consulted included Comcast, KPMG, Mastercard, Salesforce and Visa, plus the Analysis and Resilience Center for Systemic Risk, whose members include the CISOs of the largest US banks.
Under EFS, the activity data used for monitoring is stored in the customer’s own cloud account, such as Amazon S3, Azure Blob Storage or Google Cloud Storage, under the customer’s encryption keys, access policies and audit logging. Automated systems analyze a rolling window of traffic for signs of serious misuse, such as attempts to develop offensive cyber or biological capabilities or the use of stolen credentials, and send flags directly to the customer’s own reviewers, with no human review by Anthropic staff required.
EFS will roll out in phases starting this fall and will be supported on Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Google’s Agent Platform and Microsoft Foundry. Until it is ready, eligible customers get zero data retention on Fable 5 and Fable 5.1.
The design reflects a broader reality for AI vendors selling into regulated industries. Customers will accept safety monitoring, but they want custody of the data and control over who looks at it, and EFS hands both back while keeping the cross-session detection Anthropic says frontier models now need.


WebProNews is an iEntry Publication