Safe Physical AI workshop focuses on action-level assurance for embodied agents

AI safetyRobotics governanceGermany / International15–16 August 2026

Host: IJCAI/ECAI 2026 workshop at the University of Bremen. Status: completed research workshop; not a product release or consensus standard.

The programme covered verification, interpretability, uncertainty, human–robot interaction, alignment, runtime guardrails, formal methods and safe autonomy. Speakers stressed that language-model alignment alone does not guarantee safe physical action.

Why it matters and impact: The practical design signal is to put deterministic constraints, monitored recovery and human authority between model output and actuation. Workshop talks remain research claims until independently tested.