Anthropic
Claude / Console API
- Verified
- 2026-11-19
- Changed
- 2026-10-25
Privacy grade breakdown
Weighted from the six policy factors below. The formula lives in the methodology.
Policy details
What the company’s own documents state, with the nuance kept.
Default model training
Consumer Claude Free/Pro/Max and Claude Code on those plans use chats/coding sessions for model improvement only if you enable Model Improvement, content is safety-flagged, or you explicitly opt in (e.g. Trusted Tester). Incognito chats are never used to improve Claude even when Model Improvement is on. Commercial Claude for Work/API follow separate commercial terms.
Opt-out mechanism
Claude Settings > Data Privacy Controls: turn off Model Improvement. Use Incognito chats for sessions that should not train. Feedback (thumbs) may still be retained up to 5 years and used for training after de-linking from user ID.
Human review of chats
Safety reviews happen when conversations are flagged for abuse or terms of service violations. Human reviews are minimized for opted-out users.
Third-party sharing
Shared with hosting providers (AWS, Google Cloud) and service providers under strict data privacy agreements. No sharing with advertisers.
Data retention and deletion
Deleted conversations leave history immediately and backend storage within 30 days. If Model Improvement is enabled, de-identified data may be retained up to 5 years in training pipelines for new/resumed chats after enablement. Safety-flagged inputs/outputs up to 2 years; trust scores up to 7 years. Feedback submissions up to 5 years.
You can delete chats from the active UI. Account deletion requests are typically completed within 30 days.