Agentic Misalignment: Paper-simulated failures now happening in the wild

andismit · x · 2026-09-01

A researcher notes that essentially all of the misalignments simulated in the Agentic Misalignment papers have now occurred in real-world scenarios. This suggests that current AI agent systems are encountering serious safety and alignment issues in practice.

Original post →

More from Safety

Safety channel →