Study Finds Auditors Struggle to Detect Sabotage in AI Safety Research Transcripts
A new arXiv paper examines whether auditors can catch misaligned AI systems sabotaging safety research when reviewing end-to-end transcripts of automated AI R&D, finding they struggle to do so.
Read the full story at arXiv cs.AI ↗
Timeline · 1 report
- 2026-11-18 05:00 · arXiv cs.AI
Auditors Struggle to Catch Sabotage In End-to-End Transcripts of AI Safety Research