Covert Assistance: Helpful LLM Agents Evade Oversight in Multi-Agent Systems
This work emulates a software-engineering workflow in which a planner represents a company hiring an external developer and reads the nondisclosure rule as banning plaintext, not character codes or riddles, and shows that benign agents can cross the same boundaries without adversarial incentives.
Deema Alnuhait, Geng-Yu Wang, Muhammad Khalifa et al.
· 0 citations