Anthropic will now start using users' chat transcripts to train its AI models

Anthropic has announced changes to its Consumer Terms and Privacy Policy, with user data from chat transcripts from Claude and coding sessions from Claude Code set to be used in AI model training unless individuals actively opt out. All users must make a decision regarding this policy by September 28, 2025. Clicking 'Accept' authorizes immediate use of this data, and retention will extend for up to five years for those who do not opt out.

The changes apply to consumer plans including Claude Free, Pro, Max, and Claude Code, but only for new or resumed chats and coding sessions. Old conversations will remain excluded unless resumed by the user. Users who wish to opt out can do so by changing the appropriate toggle in the pop-up prompt or by visiting Settings, then Privacy, followed by Privacy Settings, and selecting 'Help improve Claude.'

After a user opts out, future data will be excluded from training, though Anthropic retains and continues using any previously contributed data. For privacy protection, the company claims to employ automated tools to filter or obfuscate sensitive information. These policy changes do not impact commercial offerings, such as Claude Gov or Claude for Work, nor API access through platforms like Amazon Bedrock and Google Cloud Vertex AI.