Anthropic on Tuesday introduced Enterprise Frontier Safeguards (EFS), a system that lets large business customers store their Claude usage logs in their own cloud infrastructure instead of Anthropic’s, while still allowing automated detection of misuse such as cyberattack attempts or the use of stolen credentials.
The company said it built EFS after enterprise customers, particularly in regulated industries such as banking and healthcare, pushed back on a data retention policy that required storing interaction logs to catch sophisticated, multi-step attacks that only become visible across many sessions and accounts.
How it works
Under EFS, activity data is written to a customer’s own cloud account — Amazon S3, Azure Blob Storage, or Google Cloud Storage — encrypted with keys the customer controls, and logged under the customer’s own access and audit policies. Automated systems still scan rolling windows of traffic for signals such as attempts to build offensive cyber or biological capabilities, according to Anthropic. Unlike the company’s earlier approach, no Anthropic staff review flagged activity directly; customers receive the flags and investigate internally.
The safeguards were shaped with input from more than 100 customers across financial services, healthcare, manufacturing, telecom, law, retail and the public sector, Anthropic said, including chief information security officers from Goldman Sachs, Morgan Stanley, Citi, Bank of America and Wells Fargo who advise through the Analysis and Resilience Center for Systemic Risk. Comcast, KPMG, Salesforce, Snowflake, Mastercard, Visa and Stripe also took part.
Rollout and cost
Anthropic plans to phase in EFS starting later this fall across Claude Code, Claude Enterprise, and Claude access through Amazon Bedrock, Google’s Agent Platform and Microsoft Foundry. Eligible customers already get zero data retention on Claude Fable 5 and 5.1 as an interim step. Anthropic said it won’t charge for EFS itself, though customers still pay their cloud provider’s standard storage and data-transfer fees.
The move follows a broader push by Anthropic to formalize how it monitors and restricts misuse of its most capable models — including the $35 million Defender Fund it launched to support cyber-defense tools built on Claude. Anthropic frames EFS as a way to expand security monitoring without asking regulated customers to hand over sensitive logs.