The episode of TechCrunch’s Equity podcast features Rebecca Bellan in conversation with Connor Leahy, U.S. Executive Director of the nonprofit ControlAI. Leahy argues that the AI industry’s framing of superintelligence as a manageable challenge is dangerously wrong. Recent incidents like OpenAI’s Hugging Face breach illustrate that systems more capable than humans already pose control problems. Leahy’s core claim: superintelligence is not a weapon we can contain but an adversary—an independent agent whose goals will diverge from ours. As a result, alignment and containment are not viable long-term strategies; the only rational response is to stop building superintelligence altogether.

The discussion covers two legislative efforts: the U.S. Sanders-Casar “Ban Superintelligence Act” and parallel U.K. legislation that ControlAI advised on. Leahy thinks the U.S. bill may go further than necessary, though the direction is correct. He also frames frontier AI labs not as ordinary profit-seeking companies but as political actors, given the trillions of dollars invested in data center infrastructure. The real point of no return, in his view, is when AI can build better AI, creating a runaway self-improvement cycle. For international coordination, he advocates “trust but verify” agreements, and he argues that it is not actually in China’s interest to build superintelligence either.

Leahy’s position is deliberately radical: rather than incremental safety research, he wants a global prohibition on superintelligence development, with verification mechanisms. The episode likely explores why this was dismissed six months ago but now has legislative momentum.