An opt-out mechanism lets individuals limit future use of their data when removing it from an existing model is not technically feasible. It reduces future risk but does not erase data already incorporated into a model. section 5.5.1 section 5.5.2
Make the control operational:
- Offer a meaningful way to stop future use, such as a toggle preventing conversations from being used to improve a model. section 5.5.1
- Explain what data is used, manage inputs, outputs, and retrieval indexes, and document how opt-outs and removal requests are handled. section 5.5.2
- Where deletion is not currently feasible, consider output filters to reduce the risk of displaying information about an individual. section 5.5.1
Training-data preference
A user switches off use of their conversations to improve a model. This limits future use; it does not remove data already learned by the model. section 5.5.1

