AgentHallu

AgentHallu is a 2026 benchmark developed to evaluate AI agent error detection, finding that even top models identify the error step only 41.1% of the time, highlighting critical reliability limitations.

Two-hop subgraph: this entity, every entity it directly relates to, and every entity those neighbors relate to. Drag a node, scroll to zoom, click to inspect — or click any neighbor and re-center the atlas there.