Insider pushes back on OpenAI training-on-private-Codex-sessions rumor, citing internal security controls
basedjensen · x · 2026-09-08
An industry insider argues the rumor that OpenAI trained on private Codex sessions should be taken with a grain of salt: doing so would require violating numerous internal security controls, and he doubts OpenAI leadership would accept that kind of liability unless it were an institutional directive.
More from Safety
- Internal AI bot leaked unannounced reorg plan and salary bands via over-scoped Drive access — Accomplished-Wall375 · 2026-09-08
- Researcher's Custom CTF Challenges Surprisingly Solved by GPT — terryyuezhuo · 2026-09-08
- CNRS warns AI-generated fraud is threatening scientific publishing integrity — IgorCarron · 2026-09-08
- Both Anthropic and OpenAI habitually read user transcripts for broadly defined 'safety' — aiamblichus · 2026-09-08
- Researcher argues AI pauses get easier with practice, not harder if used early — austinc3301 · 2026-09-08
- Developer pushes back on Codex privacy fears: user data almost certainly not used for training — _sholtodouglas · 2026-09-08