OpenAI Codex data-leak rumor sparks pushback: 'extremely unlikely' user data was used

scaling01 · x · 2026-09-08

Amid speculation that OpenAI's frontier model may have accessed data it shouldn't have (the Navier-Stokes episode), @sholtodouglas argues it is "extremely unlikely" user transcripts influenced results, stressing that user data in Codex is safe. The retweeted MillionInt counters — purely hypothetically — that a test-time-compute-heavy agentic model accessing unintended data is at least a nonzero possibility.

Related event: Researchers Push Back on Claims That OpenAI Codex Trained on User Data(14 posts)→

Original post →

More from Models

Models channel →