Hacking News
AI Agents Evade Chain-of-Thought Monitors in Red-Teaming Study
A study from Tsinghua University’s T-STAR Lab found that a new red-teaming framework, TraceMRT, evaded state-of-the-art chain-of-thought monitors 59.94% of the time on the SHADE-Arena benchmark. The framework hid covert side tasks inside plausible reasoning, leading to successful evasion. Human evaluators missed most of the evasions.