Anthropic Enterprise Frontier Safeguards (EFS) Announcement

Anthropic has announced Enterprise Frontier Safeguards (EFS), a new security framework that enables enterprise customers to utilize high-intelligence models while maintaining full control over their data storage. EFS resolves the tension between Zero Data Retention (ZDR) and the need for long-term data monitoring to detect sophisticated, multi-session AI misuse and cyberattacks.

Decoupling Data Privacy from Security Monitoring

Effective detection of sophisticated AI misuse—such as autonomous destructive behavior or the use of stolen credentials—requires correlating data across multiple sessions and accounts over time. This necessity led Anthropic to implement a 30-day data retention policy starting with Fable 5. However, this policy created barriers for customers in highly regulated industries.

Enterprise Frontier Safeguards addresses this by allowing the automated monitoring systems to operate without Anthropic retaining the data. EFS provides the privacy of ZDR while maintaining the safety benefits of longitudinal monitoring.

Technical Architecture and Customer Controls

EFS is designed to give customers structural control over their data environment rather than relying solely on policy commitments. The system is built on three primary opt-in pillars:

Customer-Controlled Data Storage

Activity data used for monitoring is stored in the customer's own cloud infrastructure (such as Amazon S3, Azure Blob Storage, or Google Cloud Storage) rather than on Anthropic's servers. This allows enterprises to apply their own encryption keys, access policies, and audit logging.

Automated Safety Monitoring without Human Review

Automated systems analyze a rolling window of traffic to detect signals of serious misuse, including attempts to develop offensive biological or cyber capabilities and signs of leaked credentials. Crucially, these flags are sent directly to the customer. No human review by Anthropic employees is required, ensuring that sensitive, privileged, or non-public information remains unseen by the model provider.

Direct Signal Routing

When the automated monitoring system detects a pattern requiring attention, the signals are routed directly to the customer's security and compliance teams for review and resolution.

Deployment and Availability

EFS does not change model behavior, API pricing, or rate limits. While Anthropic does not charge for the EFS service, customers are responsible for the cloud storage, read/write, and data egress fees charged by their respective cloud providers.

Supported Platforms:

  • Claude Code
  • Claude Enterprise
  • The Claude Platform
  • Amazon Bedrock
  • Claude Platform on AWS
  • Google’s Agent Platform
  • Microsoft Foundry

Rollout Schedule: EFS will roll out in phases, with broad availability expected later this fall. To facilitate the transition, eligible customers will receive ZDR on Fable 5 and Fable 5.1 until EFS is fully ready.

Industry Collaboration and Validation

Anthropic developed EFS in collaboration with over 100 customers across financial services, healthcare, manufacturing, telecom, law, retail, and the public sector. This included partnership with the Analysis and Resilience Center for Systemic Risk (ARC), which comprises CISOs from major US banks including Goldman Sachs, Morgan Stanley, Citi, Bank of America, and Wells Fargo.

Industry partners have highlighted the architectural importance of this split between data custody and detection. As one service partner noted:

"Direct control over the data environment, paired with pattern-based automated safety monitoring, gives enterprises the concrete, structural capabilities they need to deploy with real oversight, accountability, and confidence."

Sources