Anthropic moves the monitoring record into the customer's own cloud account

The interesting part of Enterprise Frontier Safeguards is not what it detects but where the record of it lives. Anthropic published the design on September 1, 2026: the traffic its automated systems analyze for misuse is stored in the customer's own cloud account, under the customer's encryption keys, access policies and audit logging. The analysis continues. Custody of the data changes hands.

Why there was retention to remove

Anthropic introduced 30-day data retention starting with Fable 5. The reason it gives is detection rather than training: serious misuse spreads across many sessions and many accounts, so analyzing each interaction on its own and discarding it immediately never sees the pattern. Correlating traffic over time requires keeping the traffic for a while.

That argument did not survive contact with regulated industries. Anthropic says enterprises in financial services, healthcare, law and the public sector accepted the security reasoning and still could not adopt a model that retained their data. The company says it designed the answer with more than 100 customers and with its cloud partners Amazon Web Services, Google Cloud and Microsoft Azure.

Where the data sits now

Activity data used for monitoring can be stored in the customer's own cloud account, in services such as Amazon S3, Azure Blob Storage or Google Cloud Storage. Customer-owned storage, customer-managed encryption keys and fully automated review are each opt-in, so an organization turns on only the parts it needs.

Review on Anthropic's side is automated and stops there. The systems analyze a rolling window of traffic for signals of serious misuse, including attempts to develop offensive cyber or biological capability and signs of stolen or leaked credentials. A flag goes straight to the customer, and no Anthropic employee is required to read it.

• Support is planned for Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google's Agent Platform and Microsoft Foundry.
• Anthropic charges nothing for the safeguards; the cloud provider bills storage, reads, writes and egress like any other resource.
• Turning the controls on changes no model behavior, no API pricing and no rate limits.
• Eligible customers receive zero data retention on Fable 5 and Fable 5.1 until the safeguards are ready for them.

What the announcement leaves open

The rollout is phased and the announcement names a season rather than a date, with broad availability planned for the autumn of 2026. Access is requested through a form, and Anthropic decides which customers are eligible; no published rule sets that boundary.

One thing is worth stating plainly. This changes who holds the log, not whether the traffic is examined. Automated analysis of enterprise traffic remains part of using the model, and what the organization gains is custody, keys, and seeing the flags first.