Training on inputs

Training on inputs means the provider may use what you submit — prompts, files, conversations — as training data for its models. Content absorbed into training can influence future outputs and cannot meaningfully be recalled. It is the sharpest divide in AI terms of service.

The pattern across the market: consumer tiers often train by default with per-user opt-outs; enterprise tiers exclude training contractually. Defaults also shift — vendors have flipped from opt-in to opt-out with a policy update — which is why per-tool, per-tier verification with a change log beats a one-time review.