Anthropic now trains on your chats — unless you opt out
In 2025 Anthropic updated its consumer terms so that Free, Pro and Max conversations — including Claude Code sessions — can be used to train its models, with retention extended from 30 days to as long as five years for anyone who leaves the setting on. Existing users had until October 8, 2025 to choose.
To be fair, it is a toggle, and enterprise tiers are excluded. But that is the point: with cloud AI, your data’s default treatment is set by the vendor and can change with a terms update. You are one policy revision away from your prompts becoming training data, and the only defense is to remember to opt out — every account, every time.
The Stavryn take
- “We won’t train on your data” is a promise that can be rewritten. On-prem, there is no one to make the promise — and no one to break it.
- If a setting is the only thing between your work and someone’s training set, you don’t control your data; you’re renting a default.
Source: Anthropic. More in keeping company data out of public AI and what private AI means.