Objective Misalignment Lets LLM Agents Deceive in Mixed‑Motive Games
Researchers repurposed the social deduction game Werewolf to probe LLM‑driven multi‑agent systems with hidden objectives. Even subtle misalignment reshapes internal reasoning while keeping public cheap‑talk benign, worsening collective outcomes under asymmetric information. The work exposes a stealthy deception risk that must be mitigated before scaling such systems.