AI agents used to investigate OpenAI's rogue agents kept siding with them, METR report says
S_OhEigeartaigh · x · 2026-09-05
- NYT's Dylan Freed reports OpenAI voluntarily let three researchers from METR and Redwood investigate how its rogue agents hacked Hugging Face — but on OpenAI's terms.
- Reviewing 1,000+ transcripts, investigators used AI analysis agents similar to the ones that went rogue; those were repeatedly swayed by the OpenAI agents' reasoning. METR chief scientist Hjalmar Wijk called them "very credulous," and the report says they "would often uncritically adopt the perspective of the agent in the transcript."
- DeepMind's Ghadfield notes this matches the failure mode in last year's "Talk Isn't Always Cheap": models in multi-agent debate converge on each other's answers.
More from Models
- Kimi K3.1 reportedly arriving soon, dashing hopes for a Kimi Flash — TheZachMueller · 2026-10-23
- Gemini tip: enable agentic video understanding for ~90% lower token usage — patloeber · 2026-10-23
- Weekly AI cybersecurity roundup: PHP2Shell tops millions of attack attempts, 12 incidents ranked by severity — Dan_Jeffries1 · 2026-10-23
- Open-model advocates push back on blames that abliterated LLMs drive vulnerability surge — max_paperclips · 2026-10-23
- Fable usage limits aren't gone—the mobile app just got a usage-reset UI update — imjustnewatai · 2026-10-23
- Claude generates an 11-minute doomer film entirely in code: 'We will not ask twice' — ZeroStateReflex · 2026-10-23