Anthropic is launching Enterprise Frontier Safeguards so companies can use frontier Claude models with something closer to zero data retention, without Anthropic dropping misuse monitoring.
In a Sept. 1 post, Anthropic said EFS stores activity data used for monitoring in cloud infrastructure the customer controls, not Anthropic. Customers can keep that data in their own AWS, Azure, or Google Cloud accounts, under their own encryption keys, access policies, and audit logging. When automated systems flag possible misuse, those signals go to the customer’s reviewers. No Anthropic human review is required.
That design answers pushback on 30-day retention Anthropic introduced starting with Fable 5 for Mythos-class models. Anthropic said retention was for detecting multi-session misuse, not for training on enterprise data. EFS is meant to keep that rolling-window monitoring while the logs stay in the customer’s environment.
EFS will roll out in phases starting later this fall. Until it is ready, Anthropic said eligible customers will get zero data retention on Fable 5 and Fable 5.1. The public announcement does not spell out who counts as eligible.
Anthropic said it built EFS with more than 100 customers and with AWS, Google Cloud, and Microsoft Azure, and that equivalent controls will work on Claude Code, Claude Enterprise, the Claude Platform, Bedrock, Google’s Agent Platform, and Microsoft Foundry. Customer-owned storage, customer-managed keys, and fully automated review are each opt-in. Anthropic said it does not charge for EFS; cloud providers bill for storage as usual.
This is not Anthropic walking away from safety monitoring. It is moving where the logs live and who reviews the flags.