Invited talk “Emergent Safety Risks in Multi-Agent and Single-Agent Systems” at the CVPR 2026 Workshop on Safety of Vision-Language Agents :shield: