OpenAI’s Dean Ball Urges Governments to Build a Way to Slow AI

Dean Ball, OpenAI’s head of Strategic Futures, says governments should prepare to deliberately slow AI development when necessary—while avoiding the kind of broad state authority that could threaten individual rights. Speaking as a private citizen, Ball also rejected the arms-race framing that has shaped much of the AI policy debate, saying it benefits nobody. The combined position makes pace, power and infrastructure parts of the same policy question rather than separate debates.

A brake with narrow limits

Ball’s immediate prescription is not an announced pause or a specific regulatory program. It is a call for governments to work out how AI development could be paced, so they have a usable mechanism if political leaders decide that acceleration should slow. He said the design is legally sensitive: a poorly built power could become a “hugely tyrannical” government grab over technology that may be important for speech and individual rights.

That caution reflects the premise of OpenAI’s Strategic Futures team, which Ball leads. In his August introduction to the team’s Intelligence Age blog, Ball described excessive concentration of power as its largest and most difficult long-term AI risk. The post was explicitly presented as Ball’s view, not necessarily OpenAI’s organizational position.

An arms race here benefits nobody.

Dean Ball, speaking as a private citizen

Control over systems, not only their speed

Ball locates the lasting risk less in whether software can replace a particular job than in who holds political and economic leverage after AI changes the economy. He argued that institutions should keep human preferences meaningful by preventing massive concentrations of power. That could involve questions of wealth distribution and liability, he said, but he singled out a sharper legal boundary: AI systems should not be allowed to own property.

The property boundary Ball proposes

• Explicit legal provisions preventing AI systems from owning property.
• Limits on an AI system’s ability to own real estate.
• Limits on an AI system’s ability to transform the physical world independently.

The distinction is consequential. A pacing regime addresses how fast increasingly capable models advance. Property and accountability rules address what AI-based actors can control once they operate in the economy. Ball’s position is that some form of independent AI economic activity is likely, but access to physical assets can still be constrained through law and public policy.

Europe’s different assignment

Ball’s advice for Europe runs in the opposite direction on one practical front: move faster. He said Europe should spend less effort trying to build sovereign AI systems from scratch and concentrate instead on protecting and expanding its physical-world industrial capabilities. He also said the continent needs to build data centers with private-sector leadership and far greater urgency.

That is not a call for Europe to disengage from AI policy. It is a choice of where to compete: industrial capacity and computing infrastructure rather than duplicating the model-building effort. Ball’s rationale is that Europe already has expertise making complex physical goods, while the systems being developed in the United States will require extensive real-world construction that the U.S. cannot supply alone.

The unresolved problem is institutional: who would decide when acceleration warrants restraint, and under what limits? Ball said geopolitical choices belong to elected U.S. leaders, not private companies. He assigned OpenAI a narrower role—being candid with governments about frontier developments they may not otherwise see, so officials can make informed decisions.