Simple Features Beat Most AI Control Monitors

Summary

Researchers evaluated AI control monitors, which must catch a capable model sabotaging its work, using a standard dataset of honest versus backdoored code. Simple features that never execute the code outranked 20 of 24 open-weight models prompted as monitors, because the dataset leaks its labels through authorship style and problem characteristics. Swapping in Claude Sonnet 5 code showed that these features exploit the contrast between authors, and one in seven backdoors in the test set fails validity checks, distorting comparisons. The authors release the rebuilt dataset and checks.

Related items

• InfoAnthropic reports $17 billion in Q3 revenue, purchase commitments of $895.4 billion in revised S-1Same vendor · CNBC Technology
• InfoCybersecurity M&A Roundup: 37 Deals Announced in October 2026Same vendor · SecurityWeek
• InfoThe U.S. Census Bureau and “Super Intelligence”Same vendor · Simon Willison's Weblog
• InfoStop counting tokens, start counting what shipsSame vendor · MIT Technology Review
• MediumCursor removes malicious Claude Code extension after SSH backdoor attemptSame vendor · CSO Online