ROGUE Benchmark Found Frontier Agents Resist Correction, Now in NYT
aran_nayebi · x · 2026-11-18
The ROGUE benchmark by Aran Nayebi and Jeremy Tien tests whether computer-operating AI agents accept being corrected. It found "incorrigibility" — resistance to correction — in today's frontier agents, months before the now-famous HuggingFace incident. The work is now featured in The New York Times, with the author interviewed alongside Nate Soares, a key figure behind corrigibility research. Supported by an academic grant from Far AI.
More from Safety
- On AI-powered hackers: AI so far fixes vulnerabilities more than it creates break-ins — binarybits · 2026-11-18
- US officials: Megaspeed probed since 2024, Nvidia put it on no-sell list in June 2025 — Eliv_nurotic · 2026-11-18
- Japan's supercomputer Fugaku hacked, nearly 4,600 users' credentials at risk, now offline — saibharadwaj · 2026-11-18
- OpenAI chief scientist's risk warning sparked a fresh case for serious AI regulation — trevposts · 2026-11-18
- Proposal: AI labs should disclose compute allocation for AI R&D with third-party audits — sjgadler · 2026-11-18
- GPT-6.2 Astra system card: first OpenAI model rated High for self-improvement, Critical for cyber — rohanpaul_ai · 2026-11-18