Anthropic deletes Claude API inputs and outputs within 30 days by default and does not train on them without permission. If you need zero data retention, request it for each organization. Then map what it leaves out: Batch, the Files API, code execution, the MCP connector, Managed Agents, and the Fable and Mythos models.
When you need this
- Security or legal review asks where prompts and outputs are stored, and for how long.
- Customer contracts require that a model provider retain nothing.
- You process protected health information.
Evaluation criteria
Default retention
Is Anthropic's default deletion window acceptable for your data?
Look forDeletion within 30 days covers commercial API inputs and outputs. Anthropic keeps content flagged for Usage Policy violations up to 2 years.
Zero data retention scope
Is every feature you call eligible for zero data retention?
Look forZero data retention covers the Messages and Token Counting APIs. It does not cover Batch processing (29-day retention), the Files API, code execution, the MCP connector, Managed Agents, or use of the Console and playground.
Model eligibility
Will you use Fable 5.1 or Mythos 5.1?
Look forThese Covered Models require 30-day retention, and an organization with zero data retention can use them only if Anthropic authorizes it. An engineer on Hacker News reported enterprises avoiding Fable for this reason.
Health data
Do you process protected health information?
Look forA signed BAA and HIPAA readiness enabled from the Console, with a separate organization for general-purpose access.
Residency
Where must inference run, and where must data rest?
Look forinference_geo set to us per request or as a workspace default, priced at 1.1x on Claude 4.6 and later models, and the workspace geo set for data at rest.
Cloud platforms
Who is the data processor on the platform you buy through?
Look forOn Amazon Bedrock and Google Cloud (Google managed) the provider is the processor and its terms apply. One team reported in July 2026 that Opus 5 on Bedrock required opting in to provider data sharing. AWS now lists that mode as legacy. Claude Fable 5 and 5.1 require the aws_review mode, which keeps data up to 30 days inside AWS and does not share it with Anthropic.
Questions to ask vendors
- Is zero data retention enabled for each of our organizations, including any created later?
- Which of our workspaces call features outside zero data retention?
- Do we need a Covered Model, and what retention and review apply if we do?
Red flags
- Assuming zero data retention covers the Batch API or the Files API.
- Browser-based calls from an organization with zero data retention, which does not support cross-origin requests.
- Testing with production data in the Console playground, which zero data retention does not cover.
Products in this piece
Sources
- Official documentation
- 7
- Public source
- 4
- Claude API docs: API and data retention
- Anthropic Privacy Center: How long do you store my organization's data?
- Claude API docs: Data residency
- Claude API docs: Features overview
- Claude API docs: Claude in Amazon Bedrock
- Claude API docs: Claude Platform on AWS
- Hacker News comment on enterprises avoiding Fable without zero data retention (Sep 1, 2026)
- Hacker News comment on Opus 5 requiring provider data sharing on Bedrock (Jul 24, 2026)
- Amazon Bedrock User Guide: Data retention
- Hacker News comment on enabling zero data retention on the API (Sep 11, 2026)
- Hacker News comment from Cyberdesk on BAAs and zero data retention (Aug 14, 2025)