Anthropic announced plans to change enterprise data retention policy
Anthropic building customer-owned safety-retention log system for fall 2026
Enterprise AI procurement often turns on data privacy controls and compliance requirements. A public account of a customer migrating workloads to Anthropic after OpenAI was unresponsive to data retention requests shows the commercial consequences of policy gaps.
The full picture
Anthropic is building a system, due in fall 2026, that lets enterprise customers store required 30-day safety-retention logs in their own AWS or GCP environments, with Anthropic retaining no copy. Boris Cherny, head of Claude Code, said the system has been in development with customers for some time. Anthropic maintains a 30-day retention requirement for Mythos-class models, citing safety attacks that only become visible across multiple requests. For customers using Amazon Bedrock or Google Cloud Agent Platform, Anthropic's existing documentation already shows retained data stays inside AWS or GCP; the fall rollout broadens that arrangement across other enterprise access paths. Separately, an enterprise customer publicly reported that OpenAI failed to respond to repeated Zero Data Retention requests, leading them to migrate production inference to Anthropic, which completed enterprise enablement and ZDR provisioning in the same window.
How it developed
Boris Cherny confirmed fall 2026 ship date for customer-controlled safety-retention logs in enterprise cloud environments
Sources
Want this in your inbox?
I send a short email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free