The AI agent allow list checked OpenAI against its own published data usage terms, 17 September 2026, plus the Navier-Stokes case that made this question urgent.
Every value below comes from the vendor's own published privacy policy, terms of service or terms of use, read and dated. Click show clause to see the exact sentence and its source link.
| Vendor | Category | Trains on data | Opt-out | Enterprise | API | Checked |
|---|---|---|---|---|---|---|
| ChatGPT openai.com | Code & Development | Trains unless opted out | Yes | Yes | Yes | 17 Sep 2026 |
These 7 fields were checked but their values are not shown here at all, only their names. Values are in the full register.
After an agent swarm produced a math result, two mathematicians said their own unpublished Navier-Stokes progress, developed over months with drafts fed into Codex, had reached OpenAI.
OpenAI's statement: "While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models."
OpenAI's documentation on how data is used to improve model performance places consumer Codex and ChatGPT usage under Data Controls, which train on submitted content unless a user opts out.
OpenAI's separate API data usage policy excludes API traffic from training by default, a different commitment from the consumer product.
OpenAI's Data Controls FAQ describes the opt-out as applying going forward. It does not describe retroactively erasing any training effect from sessions that ran before the setting was changed, the exact detail at issue in the Navier-Stokes case.
OpenAI's own documentation on how data is used to improve model performance describes consumer ChatGPT and Codex usage as subject to Data Controls, which train unless the user opts out, while the separate API tier is excluded from training by default.
After OpenAI announced a math result reached by an agent swarm, two mathematicians said their own unpublished progress, developed with Codex, had reached OpenAI. OpenAI's statement was that it could not rule out that de-identified data derived from their usage helped improve its models.
OpenAI's own Data Controls FAQ describes the opt-out as applying going forward rather than retroactively erasing the training effect of past sessions, which is the detail at the center of the Navier-Stokes dispute.
Yes. OpenAI's published API data usage policy states that API inputs and outputs are excluded from training by default, separately from the consumer product's Data Controls.
Seven fields, including opt-out scope, feedback exceptions and monitoring retention, are shown by name only here and unlocked in the full register.
Every field carries the date it was checked against OpenAI's published terms, shown above.
