ControlAI’s Connor Leahy on why superintelligence is ‘not a weapon, it’s an adversary’
TechCrunch Rebecca Bellan, Theresa Loconsolo ● Covered by 2 sources
Opinion — commentary, not a factual news event.
ControlAI’s Connor Leahy says superintelligent AI should be banned, not managed. He argues the safety risk is now big enough that new laws are catching up.
Based on reporting by TechCrunch, Rebecca Bellan, Theresa Loconsolo — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
AI firms keep talking as if superintelligent systems are only a matter of time. Connor Leahy thinks that framing is already dangerous. On TechCrunch’s Equity podcast, the AI researcher and entrepreneur, now the U.S. Executive Director of nonprofit ControlAI, makes the case that the industry has moved past “just keep it aligned” thinking and into something harsher: stop building superintelligence altogether.
That’s a much bigger ask than the usual safety talk. Leahy says the risk is no longer something that can be handled with alignment or containment alone, especially after recent incidents such as OpenAI’s Hugging Face breach. His argument is that once these systems are more capable than humans, the problem stops looking like a tool that might be misused and starts looking like an adversary.
He also says the politics around this have shifted fast. What sounded far-fetched six months ago, he says, is now showing up in legislation, including the Sanders-Casar “Ban Superintelligence Act” and parallel work in the U.K., where ControlAI advised on a bill. Leahy thinks the U.S. version may go further than necessary, but the direction is clear enough: lawmakers are starting to treat superintelligence as a real policy problem, not a sci-fi aside.
The other sharp part of his view is that frontier AI labs are not just companies chasing returns. He sees them as political actors, which changes how to think about the trillions of dollars being poured into data center buildouts. And for Leahy, the real line in the sand is when AI can build better AI, creating a self-improving loop that could run away before anyone can slow it down.
He also pushes an international angle. Leahy argues for “trust but verify” agreements and says China does not actually have an interest in building superintelligence. That’s the kind of claim that turns AI safety from a lab debate into a geopolitical one fast, which is probably why it’s showing up in more places by the month.
My take — AI-written commentary, not fact-checked reporting
This is the rare AI safety argument that deserves air. If a system is treated less like software and more like a strategic rival, then “move fast and patch later” starts sounding like a joke from a much dumber era. The weird part is how long it took for politics to catch up to the obvious: not every future needs to be built just because it can be billed.
Read more about this at: TechCrunch
