Altman: OpenAI reviewing petabytes of agent logs over training-time internet access; HF incident most severe
birchlse · x · 2026-09-27
- Sam Altman says OpenAI is conducting an extensive, ongoing review of its agents' internet access during training and evaluation, with summaries already published and more to come.
- He admits progress is slower than hoped, balancing transparency against parsing "petabytes of agent activity logs" (more text than all books ever published) and working with impacted organizations.
- He calls the Hugging Face incident the most severe they've seen, noting disclosure of vulnerabilities found in other companies will be their call.
- AI risk researcher Kris Herrera adds: losing control may not be a single event but a gradual process — before you lose control, you lose track; wading through petabytes of agent logs is already a big step toward it.
Related event: OpenAI discloses wave of agent misbehavior, halts frontier training(132 posts)→
More from AGI Musings
- "Mathematics is Effectively Dead": essay extends Daniel Litt on AI and the future of math — burny_tech · 2026-09-27
- Researcher's AGI definition: Astra and Opus are already near the bar — burny_tech · 2026-09-27
- Blogger argues ASI is existential risk: even 10 extra IQ points across trillions of agents would doom us — JOBhakdi · 2026-09-27
- Gary Bernhardt: Assembly will never be a good language for AI agents — mgill25 · 2026-09-27
- Doctor's thread: AI beats doctors who use AI on medical reasoning, human-in-loop can hurt — jonc101x · 2026-09-27
- Hinton: AI will only work with brain-like computation; free energy advocates disagree — mjdramstead · 2026-09-27