Anthropic’s EFS Puts Misuse-Monitoring Data in Customer Clouds, Not Its Own

Anthropic has introduced Enterprise Frontier Safeguards, a planned security layer that pairs zero data retention with automated checks for serious AI misuse. Activity data stays in cloud infrastructure controlled by the customer, while detected signals go to the customer’s personnel rather than requiring Anthropic employee review.

The rollout will begin in phases later this fall, with broad availability targeted for later in fall 2026. Eligible customers will receive zero data retention on Fable 5 and Fable 5.1 until EFS is ready, Anthropic says.

A different answer to the retention problem

Anthropic introduced 30-day data retention with Fable 5 because it says sophisticated misuse can be spread across sessions and accounts. Reviewing each interaction in isolation and discarding it immediately cannot expose that kind of pattern, according to the company. Anthropic says the retention policy was for safety monitoring, not training on enterprise data.

EFS keeps the monitoring context but shifts control of its storage. Customers can keep activity data in their own Amazon S3, Azure Blob Storage, or Google Cloud Storage accounts, under their own encryption keys, access policies, and audit logging.

The provider detects; the customer decides

The automated system examines a rolling window of traffic for signals such as efforts to develop offensive cyber or biological capabilities, and signs of stolen or leaked credentials. Flags are sent directly to the customer, whose staff can review the material; Anthropic says no human review by its employees is required.

How the controls will be sold and deployed

• Customer-owned storage, customer-managed encryption keys, and fully automated review are separate opt-in controls.
• Anthropic says EFS does not change model behavior, API pricing, or rate limits, and that it will not charge for EFS itself. Cloud providers may charge for storage, reads, writes, and data egress.
• Anthropic plans support for Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google’s Agent Platform, and Microsoft Foundry.

Anthropic says it developed EFS with more than 100 customers across industries including financial services, healthcare, manufacturing, telecom, law, retail, and the public sector. The practical question is whether customer teams can convert automated flags into reliable action as EFS reaches those environments.