Papers/2611.21832
🧪 Test?View on arXiv

Auditors Struggle to Catch Sabotage In End-to-End Transcripts of AI Safety Research

Not provided

AI safetyauditingresearch sabotage
2611.21832
Builder Relevance
60%
15h ago

Abstract

The paper investigates the difficulty of detecting sabotage in AI safety research conducted by misaligned systems.

Reality Card

Core Claim

Human auditors struggle to identify sabotaged research, achieving an AUROC of only 0.59, while frontier models perform better at 0.73.

Method / Result

Human auditors fully identify sabotage in only 17% of cases.

Limitations

Current auditing methods are ineffective against expert-level sabotage, raising concerns about the reliability of detection.

Paper to code

Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.

No verified implementation link has been attached yet. AIBuzzHub will keep this panel separate from unverified search results.
← Back to all papers