Anthropic consumer terms update: training on chats by default (2025)¶
Reference
Citation: Anthropic. "Updates to Consumer Terms and Privacy Policy", 28 August 2025 (effective 28 September 2025); Anthropic Privacy Center, "Is my data used for model training?"; Anthropic Privacy Policy, effective 8 July 2026. Type: provider terms. Links: announcement, privacy center, privacy policy.
What it is¶
A provider's own description of how its consumer and commercial tiers differ on training and retention. From 28 September 2025, data from "Claude Free, Pro, and Max plans, including when they use Claude Code from accounts associated with those plans" is used for training when a setting is on: "We will train new models using data from Free, Pro, and Max accounts when this setting is on." The setting defaults to on; users had "until October 8, 2025 to make your selection". Retention extends "to five years, if you allow us to use your data for model training", against the "existing 30-day data retention period" otherwise. The change does not apply to "Claude for Work, Claude for Government, Claude for Education, or API use, including via third parties such as Amazon Bedrock and Google Cloud's Vertex AI". The Privacy Center states: "By default, we will not use your inputs or outputs from our commercial products (e.g. Claude for Work, Anthropic API, Claude Gov, etc.) to train our models." The privacy policy states that personal data "is transferred to our servers in the US, or to other countries outside the European Economic Area", relying on standard contractual clauses where no adequacy decision applies.
Role in the record¶
- Grounds BP10 (bp10-a5): a documented case where the account tier, not the data, decides whether inputs train a model and how long they are kept. A researcher working from a personal account has accepted terms under which controlled-access data cannot lawfully be placed.
- Grounds BP11 (bp11-a2): the account tier decides training and retention. Source for the account-tier Example.
- Limits: one provider's terms at one date; terms change, and the current text should be checked before it is relied on. The record already carries this provider's grant terms through the Gagneur entry; this reference concerns the consumer terms, a separate document. Other providers' consumer tiers are not documented here.
Atom-level for/against detail and quotes are in the provenance data (assets/provenance.yml), keyed by practice atom.