Mental BoundWhat is zero data retention?
Zero data retention is a contractual and technical commitment that a provider processes your prompts, files and generated outputs only for as long as it takes to serve the request, then discards them — no content logs, no training on it. It goes a step beyond "no training on your data", which every business tier of Claude offers by default: ZDR also removes the retention window that normally exists for abuse monitoring and support.
How does ZDR work on Claude?
On the Claude API, inputs and outputs are removed within 30 days and not retained by default, and ZDR is available on request for organizations that need it; the newest "Covered Models" (currently the Fable and Mythos families) require 30-day retention and are excluded from ZDR unless expressly authorized. The same terms apply on Claude Platform on AWS and, as an option, on Microsoft Foundry. On Amazon Bedrock and Google Cloud the cloud provider is the data processor — Anthropic never sees your content — so retention is governed by AWS or Google rather than by Anthropic's ZDR program. Which arrangement you need is a governance question first; our AI governance service maps requirements to the right path. All seven ways to get Claude are compared in Seven paths to Claude.