Safe and open Physical AI.

Within the next decade, embodied AI agents will work and live among us, in our homes, workplaces, hospitals, streets, and public spaces. Their decisions will have physical consequences. We must prepare for this reality before it arrives at scale. Safety must be independently testable, with open evaluations, transparent safety mechanisms, reproducible research, and shared standards grounded in real-world conditions.

We believe critical infrastructure for Physical AI should be developed openly. Researchers, engineers, institutions, and the public must be able to inspect how systems are evaluated, understand their limitations, and help make them safer. Capability and safety must advance together: every improvement in autonomy should bring stronger oversight, robustness, failure detection, recovery, and human control.

No organization can anticipate every environment, interaction, or failure alone. Open science helps knowledge accumulate, reveals weaknesses earlier, and makes safety practices available to everyone. We are building for a future in which physical intelligence is useful, accountable, and worthy of trust. That future must be built deliberately, and it must be built together.