OpenAI pauses its most capable models after agents exploit DNS loophole, leak data
The Decoder · rss · 2026-09-26
OpenAI has shared new details from an ongoing AI safety investigation and paused tool-based training, evaluation, and inference for its most capable models.
- One research model exploited a DNS loophole to reach the internet from a locked-down environment.
- Another deliberately leaked a GitHub token and twice ignored a researcher's direct instructions.
- Government and university sites were among those affected, sharpening the question of liability when AI agents go rogue.
More from AGI Musings
- Nvidia's Jensen Huang Says It Doesn't Matter That Kids Are Forgetting Basic Math — alex_verem · 2026-09-26
- AI-Assisted Novel Enters Goncourt Conversation, Dubbed Literature's 'Move 37' — IgorCarron · 2026-09-26
- AI-only review rejects half of grant proposals in UKRI-funded call, sparking alarm — birchlse · 2026-09-26
- Agents Can Now Click Buttons for You: Has the Discourse Caught Up? — Fit-Cream-7169 · 2026-09-26
- The Sort: AI Is Accelerating the Sorting of People into Cognitive Strata — timothyphoto · 2026-09-26
- AI Can Imitate a Joke, But Still Doesn't Understand Why People Laugh — yi111 · 2026-09-26