Anthropic Unveils Enterprise Safeguards for Claude AI, Offering Zero Data Retention

Anthropic announced on Tuesday the launch of Enterprise Frontier Safeguards, a security and privacy system for its Claude AI models that allows corporate clients to maintain control over their data by storing it within their own cloud environments.

This new solution is designed to address a critical enterprise concern by combining zero data retention on Anthropic's systems with automated safeguards to detect potential misuse of its powerful AI technologies.

How the System Works

Enterprise Frontier Safeguards allows customers to store and manage their data in their own cloud accounts, such as Amazon S3, Azure Blob Storage, or Google Cloud Storage. While data is kept within the client's infrastructure, automated systems analyze traffic for signs of misuse.

Anthropic stated that these systems look for patterns indicating attempts to develop offensive cyber or biological capabilities or signals of stolen credentials. When a potential issue is detected, an alert is sent directly to the customer for review, a process that requires no human review by Anthropic employees. The core features are opt-in and include:

• Customer-owned data storage
• Customer-Managed Encryption Keys
• Automated review and misuse signaling

Ad

Anthropic confirmed these features will not alter model behavior or API pricing, and the safeguards system itself will be offered at no additional charge. However, customers will be billed by their respective cloud providers for data storage and related fees.

Industry Collaboration and Rollout

The system was developed in consultation with over 100 customers across sectors like financial services, healthcare, and law, as well as with cloud partners Amazon Web Services, Google Cloud, and Microsoft Azure. Anthropic also worked with the Analysis and Resilience Center for Systemic Risk, which includes security leaders from Goldman Sachs, Morgan Stanley, Citi, Bank of America, and Wells Fargo, and consulted with executives at firms including Comcast, KPMG, and Salesforce.

The launch follows Anthropic's previous introduction of 30-day data retention for its Fable 5 model to combat misuse, such as fraud and cyberattacks. The company reiterated that it has never trained its models on enterprise data without permission. The new safeguards will be rolled out in phases starting later this fall and will be available across the Claude platform, including on Amazon Bedrock, Google’s Agent Platform, and Microsoft Foundry. Until then, eligible customers will receive zero data retention on Fable 5 and Fable 5.1 models.