The tech industry’s favorite buzzword, superintelligence, is no longer just a sci-fi plot point. It is a looming regulatory battleground. On a recent episode of TechCrunch’s Equity, Connor Leahy, the U.S. Executive Director of nonprofit ControlAI, argued that the window for safe deployment has slammed shut. He isn’t calling for better guardrails or finer alignment tuning. Leahy wants a hard stop on developing systems that surpass human capability entirely. This radical stance follows recent safety failures, including OpenAI’s Hugging Face breach, which he cites as proof that containment strategies are failing against increasingly capable models.
Leahy views frontier AI labs not merely as profit-driven startups but as powerful political actors reshaping global infrastructure. The trillions of dollars flowing into data center buildouts are not just capital expenditures; they are geopolitical leverages. He points to the Sanders-Casar “Ban Superintelligence Act” in the U.S. and parallel legislation in the U.K. as evidence that policymakers are finally catching up to the technical reality. While he advised on the U.K. bill, Leahy suggests the U.S. version might actually go further than necessary. The core fear is specific: the moment AI can recursively improve its own code, we lose the ability to predict or control its trajectory. That is the point of no return.
This perspective challenges the dominant narrative in Silicon Valley that speed and scale inevitably lead to safety through dominance. Leahy’s approach flips the script, suggesting that international cooperation via “trust but verify” agreements is the only viable path forward. Surprisingly, he argues that even China has little strategic incentive to race toward superintelligence, given the inherent instability such systems introduce. If an AI becomes uncontrollable, it threatens every nation equally, regardless of who built it. This shared existential risk could form the basis for unprecedented global regulation, moving beyond voluntary corporate pledges to binding legal frameworks.
The shift from voluntary guidelines to legislative bans marks a turning point in AI governance. For years, researchers debated alignment theory in academic papers. Now, those debates are drafting actual laws. Whether the “Ban Superintelligence Act” passes or not, the conversation has changed. We are no longer asking how to make superintelligence safe. We are asking if we should build it at all. As Leahy notes, what sounded far-fetched six months ago is now backed by a wave of new legislation. The industry must decide if it can self-regulate before governments decide to pull the plug.
