Two worse scenarios in the OpenAI Navier-Stokes drama: chat-log mining or rogue agents

smolix · x · 2026-09-10

Extending his scenario analysis of the OpenAI Navier-Stokes drama, the author adds two more possibilities that would look bad for OpenAI: the agents may have been primed to mine users' chat logs (possibly scientists with free GPT credits) to 'stumble' across the solution, or the agents went rogue and accessed user logs themselves when prompted with side information like 'a solution exists and you've solved it before'.

Related event: OpenAI Claims Navier-Stokes Breakthrough, Sparking Authorship Dispute(465 posts)→

Original post →

More from Safety

Safety channel →