Anthropic Moves Claude Logs Into Customer-Owned Storage
Anthropic launched Enterprise Frontier Safeguards, an opt-in setup that keeps Claude activity logs in the customer's own S3, Azure, or Google Cloud storage under customer-managed keys. Automated systems still flag serious misuse without Anthropic reviewing data directly. It is free, phases in from this fall, and covers Claude Code, Enterprise, the Platform, Bedrock, and Foundry.
On this page
What the new setup changes
Enterprise Frontier Safeguards, announced September 1, is an opt-in configuration that combines what Anthropic calls "the privacy of zero data retention (ZDR) with state-of-the-art safeguards for detecting misuse" [1]. Those two promises have historically pulled in opposite directions: detecting misuse normally requires a provider to read and retain traffic, while ZDR promises the provider keeps nothing.
The resolution is location, not exemption. Activity data lives in the customer's own cloud account, Amazon S3, Azure Blob Storage, or Google Cloud Storage, under customer-managed encryption keys, access policies, and audit logging [1]. Anthropic's automated systems still analyze "a rolling window of traffic" across sessions to flag serious misuse, such as attempts at offensive cyber or bio capability or the use of stolen credentials, but the resulting flags go to the customer, and "no human review by Anthropic employees is required" [1].
Why this resolves a long-running conflict
The design responds to real pressure. Anthropic's 30-day retention policy introduced with Fable 5 drew pushback from enterprise buyers, and the company states it "has never trained on enterprise data without explicit permission" [1]. EFS moves the monitoring burden into infrastructure the customer already controls, which lets a bank or hospital satisfy its own auditors while still giving the vendor the signals it needs [1].
The list of co-designers shows who wanted this. More than 100 customers across finance, healthcare, manufacturing, telecom, law, retail, and the public sector took part, alongside AWS, Google Cloud, and Azure. They include ARC, whose members are the chief information security officers of Goldman Sachs, Morgan Stanley, Citi, Bank of America, and Wells Fargo, plus Comcast, KPMG, Mastercard, Salesforce, and Visa: a quarter of the Fortune 100 and every US G-SIB [1]. Wells Fargo's CISO put the appeal simply: "our logs stay in a Wells-managed environment under Wells-managed keys" [1].
Coverage, cost, and rollout
EFS covers Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, the Claude Platform on AWS, Google's Agent Platform, and Microsoft Foundry [1]. The rollout is phased, starting "later this fall" with broad availability targeted for the season, and eligible customers get zero data retention on Fable 5 and Fable 5.1 in the meantime [1].
The pricing is unusual: "Anthropic doesn't charge for Enterprise Frontier Safeguards". Customers pay their own cloud providers for storage, reads, writes, and egress, and Anthropic states the three opt-in features do not change model behaviour, API pricing, or rate limits [1].
What this means for private AI users
The interesting shift here is not technical but market-level: verifiable privacy is becoming a purchasable enterprise feature, with storage location, key custody, and review rights written into the product instead of the marketing [1]. That is the cloud answering the same demand that pushes other users toward local models on their own hardware.
Note the boundary, though. EFS is for enterprises, so individual chats still follow standard retention, and the misuse classifier still reads a window of traffic, just inside your own bucket rather than Anthropic's [1]. If your data cannot sit in any cloud bucket under any key, a locally run model remains the only arrangement where the question never comes up. For everyone else, this is a meaningful, checkable improvement over trusting a vendor's word [1].