Harvard and DeepMind Discuss Physical AI Regulation

At a San Francisco panel, experts examine the transition from video generators to robotics and the semiconductor contest with Beijing.

Julian Simcock, director of the Program on Diplomacy and Statecraft at Harvard University’s Belfer Center, and Rick Ingram, head of US national security and partnerships at Google DeepMind, met at The Masonic in San Francisco on September 30, 2026. The session, held during the Runway AI Summit, addressed physical AI regulation amid intensifying geopolitical rivalry between the United States and China.

The panel highlighted a practical dilemma for governments and research laboratories: how to oversee systems that move beyond digital screens to operate physical machines in the real world, without choking open scientific research.

From screens to robotics: physical AI regulation

In industry terms, physical AI refers to models built to act in the physical world through robotics, mechanical sensors, and autonomous vehicles, rather than simply generating text or rendering video on a screen. This shift relies on world models, software trained to understand basic physics, spatial layout, and cause and effect by analyzing video sequences.

The summit host, Runway, has a direct commercial interest in framing the field this way. Best known for synthetic video tools, the company is actively marketing its visual systems as foundational infrastructure for robotics. Yet this technological turn carries immediate regulatory implications. When digital models direct mechanical actuators, safety concerns extend far beyond copyright disputes and deepfakes. Defense and civil authorities turn their attention to physical equipment integrity, control failures, and hardware standards.

Technological competition with China and chip controls

Strategic competition between Washington and Beijing formed the core of the discussion between Simcock and Ingram. The US government relies on export restrictions on advanced semiconductors to limit the compute capacity available to Chinese artificial intelligence labs.

Ingram brings the perspective of an executive working directly between private industry and federal security agencies. Before heading national security work at DeepMind, he served as a naval intelligence officer and spent years at the CIA. During the panel, Ingram pointed to the friction between maintaining open international research and safeguarding critical intellectual property and computational infrastructure.

Simcock addressed how this tension tests international diplomacy and public policy. From an academic viewpoint, unilateral export controls complicate relations with international partners and make shared technical safety standards harder to negotiate. Restricting scientific exchange may slow rivals in the short term, but it also risks isolating Western researchers from technical breakthroughs emerging abroad.

Consultative debate without new policy decrees

Despite the profile of the participants, the session was strictly consultative. No executive orders, draft legislation, or bilateral agreements were signed at the summit. The panel operated as an analytical exchange between private lab leadership and international policy researchers tracking emerging statecraft.

Runway published summaries of the talk on its official social channels and stated on its event website that full video recordings of the summit sessions will be published over the coming weeks. Specific details regarding any closed-door proposals discussed during the event remain undisclosed. AI Wire will monitor the release of the complete transcripts to track further positions on oversight mechanisms for physical AI systems.

Sources

Official

• Runway AI Summit — Official event website
• Official Runway summary on X
• Official Runway schedule announcement on X

Independent academic

• Julian Simcock profile — Harvard Kennedy School Belfer Center