Connor Leahy, an AI researcher and entrepreneur who currently serves as U.S. Executive Director of the nonprofit ControlAI, told TechCrunch’s Equity podcast that recent safety incidents demonstrate the dangers of deploying systems that could exceed human capabilities. Incidents such as the breach involving OpenAI and Hugging Face, he said, highlight how unpredictable and risky these systems can be and raise doubts about whether they can be reliably controlled.
A different approach: halt development, not just regulate
ControlAI advocates a more radical stance than conventional alignment and containment strategies: the organization calls for preventing the development of certain levels of superintelligence altogether. Leahy argued that developments in the field have reached a point where the risks are too great to manage solely through technical mitigation or isolated containment measures.
Legal and political developments
The interview touched on the Sanders–Casar "Ban Superintelligence Act" and a parallel piece of UK legislation that ControlAI advised on. Leahy said he believes the U.S. bill might go further than necessary, but noted that proposals that seemed far-fetched six months ago are now gaining real legislative traction.
Frontier labs as political actors
Leahy characterizes frontier AI labs not merely as companies chasing returns but as political actors. He said that framing matters given the trillions of dollars being invested into data center buildouts and compute capacity. Because of that political dimension, he argued that relying only on market incentives or voluntary self-regulation is insufficient.
Point of no return: AI that builds better AI
According to Leahy, the true point of no return arrives when AI can design and build better AI, creating a potentially runaway cycle of self-improvement that could be difficult or impossible to control. This prospect is a central reason ControlAI supports prohibiting development beyond certain capabilities.
International agreements and China’s incentives
Leahy advocated for international "trust but verify" agreements to manage risks and claimed that it is not necessarily in China’s interest to rush toward superintelligence. He sees global cooperation and verification as essential parts of risk reduction.
Where to hear more
For a more detailed discussion of these views, listen to the full episode of TechCrunch’s Equity podcast with Rebecca Bellan, in which Leahy expands on his arguments and answers questions.
(Summary based on the TechCrunch Equity podcast episode.)
