Szepesvári: OpenAI's off switch stops training on your chats, but data retention remains
CsabaSzepesvari · x · 2026-09-16
RL scholar Csaba Szepesvári clarifies OpenAI's data settings: with "improve the model for everyone" off, OpenAI does not train on your chat logs (he recalls official confirmation). But the careful wording concerns data retention — models can still use past conversations in-context to "improve your experience," so off ≠ data gone. Context: community scrutiny over OpenAI's handling of Navier-Stokes conjecture chats.
Related event: DeepMind Researcher Questions OpenAI's Data Privacy Settings(4 posts)→
More from Safety
- AI's greatest risk isn't rogue robots: treat agent failures like defective products — Classic-Acadia272 · 2026-09-17
- Intel engineer pushes back on Sky News claim that AI lacks regulation — alexvoica · 2026-09-17
- AI labs want in-house auditors — but maybe they should shut the front door first — TechCrunch AI · 2026-09-17
- OpenAI quietly updates its privacy policy — LuizaJarovsky · 2026-09-17
- Canada and Germany pledge up to $300M to Bengio's safety lab LawZero — RishiBommasani · 2026-09-17
- House's No.4 Democrat Ted Lieu pushes committee vote on bipartisan FRONTIER AI bill — Miles_Brundage · 2026-09-17