Anthropic has introduced Enterprise Frontier Safeguards (EFS), a new service that lets approved corporate customers use zero data retention (ZDR) while still allowing the company to monitor for misuse of its top-tier Claude models. The move responds to complaints from enterprise clients, particularly in regulated industries, who found it difficult to adopt Anthropic's models after the firm said earlier this year it would temporarily retain prompts and outputs from its flagship models to check for safety violations, even for customers with existing ZDR agreements. The shift suggests Anthropic sees non-zero data retention as a competitive disadvantage against rivals such as OpenAI, which recently launched a similar compliance-friendly monitoring tool.

Under standard terms, Anthropic keeps commercial API inputs and outputs for 30 days unless another arrangement is in place; EFS instead stores data in infrastructure controlled by the customer rather than Anthropic, with automated monitoring alerts sent directly to the customer for review, effectively shifting some safety-checking responsibility onto enterprise users. ZDR is not automatic and must be applied for. EFS will be available by invitation this autumn across Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Google's Agent Platform and Microsoft Foundry, but only for one of the two new model versions; the other remains under the previous retention policy. Consumer-tier Claude plans are unaffected and continue under existing data retention rules.