AM-Bench: A theoretically-grounded benchmark for agentic misalignment

This is a brief overview of our new paper (📄) AM-Bench: A Unified Taxonomy and Evaluation Suite for Agentic Misalignment accepted at NeurIPS E&D 2026 💫OverviewWe introduce AM-Bench, a unified taxonomy and evaluation suite for agentic misalignment, a failure mode where models actively choose harmful actions over failure recently revealed by Anthropic. We ground our custom, realistic 35 scenarios spanning 7 different industries including software and healthcare in the fraud triangle, a criminolo...