CHI-Bench: Can AI Agents Automate End-to-End, Long-Horizon, Policy-Rich Healthcare Workflows?
$ ext{CHI}$-Bench tests AI agents' ability to automate complex, long-horizon healthcare tasks like prior authorization and utilization management. The benchmark stresses policy density and multi-role composition using a massive knowledge base. Findings show current agents struggle significantly with clinical reasoning and policy compliance, indicating a need for more robust grounding in complex medical rules.