Challenged the characterization
“Did any human or agent look at user data as part of the Navier Stokes effort? No.”
Separates a denial of targeted access during the project from acknowledged use of de-identified data and feedback to improve products.
- Speaker
- Mark Chen
- Capacity
- OpenAI researcher explaining the company's process on his account
- Observed
- September 10, 2026
- Published here
- September 10, 2026
- Targets
- OpenAI
- Classification rationale
- Narrow factual counter-position; does not eliminate the company's caveat about general training data.
Surrounding material
Full context
Before the quotation
Addresses public confusion over the official statement's data caveat.
After the quotation
This distinction does not resolve the separate normative question of whether training on research supplied by users is appropriate.
Evidence
Complete source record
Mark Chen distinguishes targeted user-data access from model improvement (opens in a new tab)
Explicit denial that humans or agents accessed user data for the effort, while affirming broader use of de-identified data and feedback.