Summary

Design document for multi-modal attack scenarios targeting Vision-Language-Action models. Extends text-only attack families to include visual and action-space manipulation vectors specific to embodied AI systems.