Training opt-out
Also: opt out of AI training, AI data training toggle
A training opt-out is the setting, sometimes a toggle and sometimes just a plan upgrade, that stops an AI vendor from using your prompts, uploads or conversations to train or improve its models.
Many consumer AI products train on user activity by default: your prompts and the model's responses may be reviewed by humans and used to improve future model versions, unless you take an explicit action to stop it. A training opt-out is that action, a setting in your account (often called something like 'Improve the model for everyone' or 'AI data retention') that, once disabled, excludes your data from that pipeline going forward.
The mechanism varies by vendor
The mechanism differs a lot by vendor and by plan. Some tools give consumers a real, self-service toggle. Others make training exclusion an automatic feature of paid business or enterprise tiers rather than something you switch off yourself: you get the exclusion by being on the right plan, not by finding a setting. And a few products do not offer an opt-out at all on any tier, meaning training use is simply a condition of using the product.
Two caveats worth knowing
Two caveats matter. First, opting out is rarely retroactive: it usually stops future data from being used, without necessarily removing what was already ingested previously. Second, 'opt-out' and 'opt-in' are opposite defaults with very different risk profiles: opt-out means you are being trained on unless you act, opt-in means you are safe by default unless you deliberately choose to contribute data. Always check which direction a given tool defaults to.
The tier gap again
Business and enterprise tiers of major assistants, such as ChatGPT Business, Enterprise and API, Claude's commercial and API access, Microsoft 365 Copilot, and GitHub Copilot Business and Enterprise, are generally excluded from training by default, in contrast to their free consumer equivalents. That gap between tiers is exactly why 'does this product train on your data' needs a tier-specific answer, not a single yes or no for the whole brand. The no-training compliance hub and the AI Tool Risk Directory record each tool's opt-out mechanism and default.