Partnering with Customers to Build Enterprise Frontier Safeguards

Meta description: Anthropic announces Enterprise Frontier Safeguards, combining zero data retention with misuse detection, rolling out to customers in phases starting later this fall.

Tags: Anthropic, Enterprise Frontier Safeguards, Claude, data privacy, AI safety

Anthropic announced Enterprise Frontier Safeguards (EFS) on Sep 1, 2026, a solution that pairs zero data retention (ZDR) with advanced safeguards for detecting misuse. The core design principle: customer data stays in cloud infrastructure controlled by the customer, not Anthropic. EFS begins phased rollout later this fall, and eligible customers receive ZDR on Fable 5 and Fable 5.1 during the transition.

The problem EFS was built to solve

Mythos-class models such as Claude Fable 5.1 represent a significant jump in intelligence and agentic capabilities, which also raises the potential for misuse and autonomous misbehavior. Anthropic says it has seen substantial evidence of attempted misuse in recent months, from fraud to sophisticated cyberattacks in which agents autonomously engage in destructive behavior. Some cases involve stolen enterprise credentials, which are hard to detect without monitoring traffic and flagging abnormal behavior.

Because the most sophisticated misuse spans multiple sessions and accounts, analyzing each interaction separately and discarding data immediately does not work. Effective detection requires storing data long enough to correlate across time and accounts. That is why Anthropic introduced 30-day data retention starting with Fable 5. The company states this policy was not motivated by training on enterprise data; Anthropic says it has never trained on enterprise data without explicit permission and never will.

The tension: enterprises in regulated industries understood the security value of retention but struggled to use models that retained data. EFS was designed with customers to deliver both ZDR privacy and cross-time monitoring.

How it was designed and who shaped it

Anthropic developed EFS with more than 100 customers across financial services, healthcare, manufacturing, telecom, law, retail, and the public sector, plus cloud partners Amazon Web Services, Google Cloud, and Microsoft Azure. Conversations spanned a quarter of the Fortune 100, every US global systemically important bank, and virtually every regulated industry.

Participants included the Analysis and Resilience Center for Systemic Risk (ARC), whose members include CISOs of major US banks such as Goldman Sachs, Morgan Stanley, Citi, Bank of America, and Wells Fargo. Anthropic also worked with leaders at Comcast, KPMG, Mastercard, Salesforce, and Visa.

What customers get

Data storage. Customers store activity data in their own cloud accounts, including Amazon S3, Azure Blob Storage, or Google Cloud Storage, under their own encryption keys, access policies, and audit logging.

Monitoring. When automated systems detect a pattern needing attention, signals go directly to the customer for review. No Anthropic human review is required. Automated systems analyze a rolling window of traffic for signals including attempts to develop offensive cyber or biological capabilities and signs of stolen or leaked credentials.

Human review. In regulated industries, review must be handled by the customer's own cleared personnel, who already work under rules governing privileged legal material, non-public information, and drug-safety reports.

EFS will be supported on Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google's Agent Platform, and Microsoft Foundry.

What to watch

The phased rollout later this fall will test whether architecture-level controls, rather than policy commitments alone, can move regulated enterprises from experimentation to deployment at scale. Customer statements from Wells Fargo, Comcast, KPMG, Mastercard, Salesforce, Visa, Snowflake, Stripe, FIS, Cognition, and Rogo emphasize the same split: customer custody of data, Anthropic operation of detection.