Anthropic’s Enterprise Frontier Safeguards keep Claude logs in the customer’s cloud

Anthropic on 1 September 2026 announced Enterprise Frontier Safeguards (EFS), a product for teams that want frontier Claude models without sending monitoring data to Anthropic’s own storage. The company describes EFS as combining “the privacy of zero data retention (ZDR) with state-of-the-art safeguards for detecting misuse.” Data used for monitoring is stored in cloud infrastructure the customer controls, not at Anthropic.

This is a separate announcement from the same day’s Fable 5.1 / Mythos 5.1 model launch. EFS is about where logs live and who reviews them when agents run on Claude, including Claude Code.

Why Anthropic says retention exists

Anthropic said it introduced 30-day data retention starting with Fable 5 because serious misuse can span many tasks, sessions, and accounts. “It is not sufficient to run automated analysis on each interaction separately and then instantaneously discard the data,” the company wrote. It also said the policy was not for training: “Anthropic has never trained on enterprise data without explicit permission, and never will.”

That retention collided with regulated buyers. Anthropic said enterprises understood the security case but “many—especially in regulated industries—found it difficult to use models with data retention.” EFS is the company’s answer: keep the monitoring window, move the bytes.

The company said it designed EFS with more than 100 customers in financial services, healthcare, manufacturing, telecom, law, retail, and the public sector, plus Amazon Web Services, Google Cloud, and Microsoft Azure. It named working groups including the Analysis and Resilience Center for Systemic Risk (ARC) and conversations spanning “a quarter of the Fortune 100” and “every US global systemically important bank.”

How EFS is supposed to work

On the Anthropic page, EFS has three opt-in pieces: customer-owned storage, customer-managed encryption keys, and fully automated review with no Anthropic human review required. “None of them change model behavior, API pricing, or rate limits.”

Activity data for monitoring can sit in the customer’s own cloud account, such as Amazon S3, Azure Blob Storage, or Google Cloud Storage, under the customer’s keys, access policies, and audit logging. When automated systems flag a pattern, “those signals are sent directly to customers so they can review what the automated systems detected.” Anthropic said flags can include attempts to develop offensive cyber or biological capabilities and signs of stolen or leaked credentials; the customer’s own people take it from there.

Support is listed for Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google’s Agent Platform, and Microsoft Foundry. Anthropic said the same controls are meant to apply whether a customer hits Claude directly or through a cloud partner, and that it is working to support eligible third-party offerings as well.

Anthropic does not charge for EFS. If a customer stores data in their cloud account, their provider bills storage, reads, writes, and egress “the same way it bills any other resource.”

Quoted customer comments on the announcement page include Wells Fargo CISO Munish Kumar Sharma, who said logs stay in a “Wells-managed environment under Wells-managed keys,” and Stripe’s Matthew Kemelhar, who said conversation logs would remain in Stripe’s AWS environment under Stripe’s security controls. Cognition CEO Scott Wu and Factory CTO Eno Reyes are among the coding-agent vendors Anthropic quoted on keeping customer code and identities on the customer side.

Timing for Claude Code shops

EFS “will be rolling out to customers in phases, starting later this fall,” with a goal of broad availability later this fall. Anthropic is taking access requests via a form on the announcement. Until EFS is ready, “eligible customers will receive ZDR on Fable 5 and Fable 5.1.”

For teams already running Claude Code or Claude via Bedrock, Foundry, or Google’s Agent Platform, the 1 September post is the source of record: customer-held monitoring data, no extra Anthropic fee, no change to model behavior or rate limits, and a fall phased rollout rather than a same-day switch.

Source: Anthropic, 1 September 2026.