Does Claude Code (Consumer: Free / Pro / Max) train on your code?
Anthropic · Consumer: Free / Pro / Max · checked 2026-10-19
Your choice, but say yes and your sessions are kept for 5 years. Check the setting.
Trains on code
Choice at signup
Default
New users pick at signup; existing users had to choose by Oct 8, 2025. Anthropic doesn't say which way the toggle is pre-set. If allowed, sessions (incl. Claude Code) are used to train new models.
How to opt out
claude.ai/settings/data-privacy-controls > turn off the model-improvement setting. Also: don't send /feedback or /bug transcripts (kept 5 years); answer 'No' to the transcript-share survey or set CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY=1.
Retention
Training allowed: 5 years. Training off: 30 days. /feedback, /bug, /share transcripts: 5 years. Survey-shared transcripts: up to 6 months. Local transcripts in ~/.claude/projects for 30 days (cleanupPeriodDays).
Telemetry
Yes: metrics (no code, prompts or file paths) default on; error reports on for Pro/Max sign-ins (v2.1.198+); session-quality surveys default on.
Disable telemetry
DISABLE_TELEMETRY=1, DISABLE_ERROR_REPORTING=1, DISABLE_FEEDBACK_COMMAND=1, CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY=1, or all at once with CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 (the WebFetch hostname check needs skipWebFetchPreflight: true).
Zero data retention
no · ZDR is only for qualified Claude for Enterprise orgs.
Offline / local model
no · Anthropic documents Anthropic API, Bedrock, Google Cloud, Foundry and LLM gateways, not local models.
Self-hostable
no · CLI runs locally, model is hosted.
Score
33/100 · training 20/40 · retention 3/25 · offline 0/20 · telemetry 10/15
Classes: training Choice at signup · retention Years · telemetry On, can disable · offline no
Not published (not found in the vendor's official sources)
default position of the training toggle at signup