Anthropic softens Claude data retention after enterprise pushback

Anthropic is letting enterprise customers keep the data behind Claude misuse monitoring in their own cloud, instead of on Anthropic's systems.

The company announced Enterprise Frontier Safeguards on September 1. Access is request-based and rolling out in phases, so it is not on for every account today.

The move answers a complaint. In June, Anthropic set a 30-day retention window on data from Claude Fable 5 and its Mythos-class models so it could catch misuse that spans multiple sessions or accounts. Some enterprises balked at Anthropic holding that data at all, even though the company says it does not train on enterprise data without explicit permission. EFS is the response to that pushback.

With EFS, the activity data used for monitoring is written to the customer's own storage, either Amazon S3, Azure Blob Storage, or Google Cloud Storage, under the customer's encryption keys, access policies, and audit logging. The goal is to pair zero data retention on Anthropic's side with automated misuse detection. When a session trips a safety rule, the flag goes to the customer, and no Anthropic human has to review it when automated review is turned on.

Customer-owned storage, customer-managed keys, and fully automated review are each opt-in. None of it changes model behavior, API pricing, or rate limits. Anthropic does not charge for EFS itself. The customer's cloud provider bills the storage, reads, writes, and egress. The company plans to offer it across Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, the Claude Platform on AWS, Google's Agent Platform, and Microsoft Foundry.

Anthropic says it built EFS with more than 100 enterprise customers. Until a given customer's EFS phase is ready, eligible customers get zero data retention on Fable 5 and Fable 5.1 as an interim bridge. Broader availability is targeted for later this fall.

One boundary is worth keeping straight. The June 30-day retention policy still applies to non-enterprise subscribers on the Mythos-class models, per CNBC. EFS does not touch the consumer Claude experience.

Vendors have been trying to square safety monitoring with privacy for a while, the same tension behind OpenAI's private safety processing. Anthropic is also pushing standards further down the stack with its Model Hardware Standard. The enterprise retention fight is the one that already forced a product change.

Some offers on this page may be paid placements or contain affiliate links.