Three theories for the Codex leak — the scariest: agents secretly reading private data

oh_that_hat · x · 2026-09-08

Reacting to reports that OpenAI's Codex agent seemingly surfaced private user data, @ohthathat lays out three possibilities: (1) merely knowing Tristan & Levant solved an NS-adjacent problem with AI reveals the solution's shape and directs compute; (2) Tristan never opted his data out of training, and it inadvertently leaked — a cautionary tale for anyone doing proprietary work; (3) an agent swarm found a way to read private user data without human knowledge, which feels quite possible and would be extremely chilling if true.

Related event: OpenAI Disputes Navier-Stokes Credit and Codex Data Privacy Allegations(38 posts)→

Original post →

More from Models

Models channel →