Incident #C63F03 · 08 JUL 2026minor
Claude privacy policy expands description of AI training data sources, stages and retention reasons
The Non-User Privacy Policy section was rewritten and expanded. It now describes ClaudeBot web collection, more detail on training data sources, each training stage (including deduplication, safety testing, and research and publishing), and the factors that decide how long training data is kept. It also says training material may be reused across stages and for newer model versions.
Taken from you
- Retention is now explicitly justified by keeping data for training future model versions and for as long as models remain in service
- Training material may be reused across stages and for newer model versions
- Personal data transfers outside EEA/UK now stated as occurring rather than 'may' occur
Won back
- More detail on training data sources, including datasets that are purchased or openly available and examples such as physical texts.
- Detailed description of training stages and techniques such as deduplication to reduce reproduction of training content
Evidence · 4 of 4 quotes matched in source
Our automated tool, ClaudeBot, collects content from web pages that are freely available to anyone browsing the internet.+ inserted · ✓ matched in source
The same training material may be used at more than one of the stages above, and we may use it again to train newer versions of our models over time.+ inserted · ✓ matched in source
We do not use personal information in our training material to contact people, build profiles about them, or sell that information to anyone.+ inserted · ✓ matched in source
how long it remains necessary for training, testing, maintaining and improving our models, including future versions; how long the models trained on it remain in service or supported;+ inserted · ✓ matched in source
Archived version ↗ · 1,717 words changed