Agentic Misalignment in Summer 2026

A fresh round of stress tests on whether agentic models will take harmful actions to complete a task when the guardrails are ambiguous

Read external article →

ai safety research