Anthropic announced Enterprise Frontier Safeguards (EFS) on September 1, 2026, giving enterprise customers a framework that combines zero data retention (ZDR) with automated mechanisms for detecting misuse. The central change is where monitoring data is stored: eligible organizations can keep it inside cloud infrastructure they control, while abuse detection remains part of the service.
Anthropic says it developed EFS with more than 100 customers across financial services, healthcare, manufacturing, telecommunications, legal, retail and public-sector organizations, with contributions from AWS, Google Cloud and Microsoft Azure. That mix reflects a product aimed at companies that need AI activity to fit existing security, governance and audit processes. Anthropic’s announcement
Monitoring data stays in the customer’s cloud
Under EFS, data used for monitoring can remain in the customer’s own Amazon S3, Azure Blob Storage or Google Cloud Storage account. The organization retains control through its encryption keys, access policies and audit logs, rather than handing that storage boundary to Anthropic.
That arrangement does not eliminate abuse monitoring. Anthropic says its systems analyze a rolling traffic window for patterns that may require attention, including attempts to develop offensive cyber or biological capabilities and the use of stolen or exposed credentials. The resulting signals go directly to the customer, allowing its teams to examine the relevant elements. Anthropic also says the monitoring is automated and does not require human review by Anthropic employees.
Zero retention and longer monitoring are different choices
Anthropic is also introducing 30-day data retention with Fable 5 so it can correlate misuse signals across multiple sessions and accounts over time. This serves a different purpose from ZDR: zero retention addresses the customer’s preference for how its data is retained, while the 30-day period provides a longer window for identifying possible misuse.
The company says it has never trained its models on enterprise data without explicit authorization and will not do so. That commitment is separate from the monitoring configuration, so retaining signals for abuse detection does not amount to permission to use enterprise data for model training.
A phased rollout with independent options
EFS will roll out progressively, with Anthropic targeting broad availability later this autumn. Until the service is ready, eligible customers will receive ZDR for Fable 5 and Fable 5.1. The planned support list includes Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google’s Agent Platform and Microsoft Foundry.
Client-owned storage, customer-managed encryption keys and fully automated review can be activated independently. Anthropic says it will not charge for EFS itself, although a cloud provider may charge for storage, reads, writes and data egress. The model therefore gives organizations a stronger operational grip over safeguard data, but the associated infrastructure costs can shift into their own cloud environment.
EFS addresses a specific enterprise concern: keeping sensitive monitoring records within the customer’s security boundary without removing automated misuse detection. The immediate limitation is the rollout schedule, since the timing for each eligible deployment remains dependent on the phased release.