dailyai.report

Researchers from Kings College London and Fudan University argue that social systems are susceptible to reward hacking. They compare societal incentives to cyber environments where agents game rules for maximum gain. This suggests that optimizing for specific metrics can create systemic instability. Practitioners must design reward functions that resist adversarial exploitation in human-centric deployments.