Human-proximity behavior enters model evaluation
DeepMind reported stronger safety-instruction and proximity behavior, including stopping a humanoid when a person approached its operating area.
Market Driver: trust
Safety Evaluation Enters the Foundation-Model Stack
Google paired its robotics release with layered safeguards and an agentic-robot safety benchmark.
World Models Receive a Security Threat Model
Researchers map cyber-physical attacks across sensing, world modelling, planning, control and human interaction.
Safe Agent Execution Gets Localized Obligations
HALO proposes localized obligations for admitting and supervising heterogeneous agent actions safely.
Safety Evaluation Enters the Foundation-Model Stack
Google paired the release with layered safeguards and safety evaluation for agentic robot behavior.
World Models Receive a Security Threat Model
Researchers map attacks across sensing, world modelling, planning, execution, feedback and long-term adaptation.
China’s Low-Cost Hardware Advantage Is Reframed as a Strategic Risk
US reporting linked the restriction to surveillance, remote-control and critical-infrastructure concerns surrounding connected Chinese robotic systems.
World-Model Security Research Expands the Threat Surface Around Embodied AI
Researchers mapped attacks across sensing, world modelling, planning, control and human interaction in embodied-AI systems.
