Anthropic Lets Enterprises Keep Claude Logs in Their Own Cloud

Who holds the misuse logs

Anthropic spent the summer telling banks it had to keep their prompts. On Tuesday it told them they can keep the prompts themselves.

The split

The product is Enterprise Frontier Safeguards, announced 1 September. Anthropic’s automated systems still watch a rolling window of traffic for serious misuse — offensive cyber, biological capabilities, stolen credentials — but the activity data lives in the customer’s own cloud account, under the customer’s keys, access policies and audit logs. When something trips, the flag goes to the customer’s people. No Anthropic employee has to look. Zero-data-retention privacy, the company says, plus the correlation across sessions that instant discard cannot do.

Why they kept the logs

That last point is why this exists. With Fable 5 in June, Anthropic imposed 30-day retention on its most capable models, arguing that sophisticated abuse does not fit in a single chat. It pledged the data would not be used for training. Enterprises, especially in regulated industries, still balked at adding another “trusted data vendor.” Kate Jensen, head of Americas, told CNBC the company spent hundreds of hours on a workaround: “We can do the scanning that we really need to… but it’s on their systems and their data so that they don’t have to compromise on their own privacy posture.”

Who asked for it

The design was workshopped with more than 100 customers — Anthropic says a quarter of the Fortune 100, every US global systemically important bank, and groups such as ARC, whose members include the CISOs of Goldman Sachs, Morgan Stanley, Citi, Bank of America and Wells Fargo. Munish Kumar Sharma, Wells Fargo’s CISO, put the bargain in one sentence: logs in a Wells-managed environment, Wells-managed keys; Anthropic operates the detection.

Price, terms, timing

EFS is free from Anthropic. The customer pays AWS, Azure or Google for storage, reads, writes and egress. Customer-owned storage, customer-managed keys and fully automated review are each opt-in; none of them change model behaviour, API price or rate limits. Support is planned across Claude Code, Claude Enterprise, the Claude Platform, Bedrock, Google’s Agent Platform and Microsoft Foundry. Rollout is phased, aimed at later this fall. Until then, eligible customers get zero retention on Fable 5 and 5.1. CNBC notes the 30-day rule still applies to non-enterprise Mythos-class users.

Privacy at the frontier is no longer a policy FAQ. It is a product feature, and a custody fight.

This account is based on Anthropic’s 1 September announcement and same-day reporting by CNBC. Customer quotations are from the company post.