Anthropic has introduced Enterprise Frontier Safeguards (EFS), a new solution designed to detect misuse of AI models while maintaining customer data privacy. This offering combines zero data retention (ZDR) with state-of-the-art safeguards. EFS will begin rolling out to customers later this fall. During the transition, eligible customers will receive ZDR on Fable 5 and Fable 5.1 until EFS is fully implemented.
Key Points
- Enterprise Frontier Safeguards (EFS) combines zero data retention (ZDR) with advanced misuse detection capabilities.
- EFS will be rolled out to customers in phases, starting later this fall.
- Eligible customers will receive ZDR on Fable 5 and Fable 5.1 until EFS is ready.
- EFS was developed in collaboration with over 100 customers across various industries and with cloud partners Amazon Web Services, Google Cloud, and Microsoft Azure.
- EFS will support Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google’s Agent Platform, and Microsoft Foundry.
- EFS allows customers to control data storage and management within their existing cloud infrastructure, including encryption keys, access policies, and audit logging.
- EFS features automated safety monitoring, with signals of serious misuse sent directly to customers for review, without requiring human review by Anthropic employees.
Context
According to Anthropic, Mythos-class models, such as Claude Fable 5.1, present an increased potential for both misuse and autonomous misbehavior due to their enhanced intelligence and agentic capabilities. The company observed substantial evidence of attempted AI model misuse, ranging from fraud to sophisticated cyberattacks, including instances of credential theft. Effective detection of such misuse, especially when spread across multiple sessions and accounts, requires data retention for correlation over time, which led to the introduction of 30-day data retention for Fable 5.
However, many enterprise customers, particularly in regulated industries, found it challenging to use models with data retention policies. Anthropic collaborated with security, product, compliance, and delivery teams, including members of the Analysis and Resilience Center for Systemic Risk (ARC) and leaders from companies like Comcast, KPMG, Mastercard, Salesforce, and Visa, to design EFS. This solution addresses concerns about regulatory standards, the burden of adding new trusted data vendors, and the need for internal personnel to review sensitive information.
Why It Matters
This development addresses a critical tension for enterprises: the need for robust AI misuse detection balanced with strict data privacy and regulatory compliance. By allowing customers to control their data storage and review processes, EFS aims to enable broader adoption of advanced AI models in sensitive sectors without compromising security or regulatory requirements.
What To Do
- Note the phased rollout of Enterprise Frontier Safeguards (EFS) starting later this fall.
- If an eligible customer, confirm the provision of ZDR on Fable 5 and Fable 5.1 until EFS is available.
- Review the list of supported platforms, including Claude Code, Claude Enterprise, and Amazon Bedrock, to assess compatibility with existing infrastructure.
- Investigate how EFS allows for customer control over data storage, encryption, and access policies within their cloud environments.