Anthropic promises zero data retention – but customers must check it worked
AI-summarised brief · reviewed before publication
Anthropic is introducing Enterprise Frontier Safeguards (EFS) for its Fable 5.1 and Mythos 5.1 models, promising zero data retention for enterprise customers. Unlike standard commercial API users, who face thirty-day input and output retention, enterprises must apply for this privacy feature. Previously, Anthropic retained data temporarily to detect safety violations and model abuse, citing substantial evidence of misuse attempts. This practice drew criticism from regulated industries uncomfortable with any data storage. EFS addresses these concerns by shifting safety processing to customer-controlled cloud infrastructure, ensuring Anthropic does not store the data. The service aims to balance security needs with strict privacy requirements. While Anthropic maintains it never uses enterprise data for training, the new system allows companies to verify compliance. This move responds to competitive pressure from rivals like OpenAI, which recently launched similar privacy-focused safety tools. The initiative seeks to make Fable more appealing to large corporate clients wary of data exposure.
💡 Why It Matters
- · By moving safety checks to customer-controlled infrastructure, Anthropic resolves the conflict between regulatory compliance and AI security monitoring.
- · This architectural shift allows enterprises to maintain strict data sovereignty while still benefiting from automated abuse detection.
