The tech industry’s favorite buzzword is superintelligence. Executives treat it as an inevitable horizon, a destination we are racing toward with breathless anticipation. But Connor Leahy isn’t buying the hype. He sees a different reality. Recent security failures, including the breach involving OpenAI and Hugging Face, prove that even current systems are difficult to contain. If we can’t secure today’s models, why do we think we can control something smarter than us?
Leahy, now the U.S. Executive Director of the nonprofit ControlAI, argues that traditional safety measures like alignment are insufficient. You can’t just tweak the code and hope for the best. His organization pushes for a radical shift: stopping the development of superintelligence entirely. This isn’t about slowing down. It’s about hitting the brakes. He joined TechCrunch’s Equity podcast to explain why the risks have outpaced our ability to manage them through technical fixes alone.
Six months ago, this stance might have sounded like fringe alarmism. Today, it has legislative momentum. Leahy points to a growing wave of new laws that back his cautious approach. The conversation is shifting from how fast we can build to whether we should build at all. This marks a significant departure from the dominant Silicon Valley narrative. It challenges the core assumption that progress is always good if it’s fast enough.
The implication is stark. If ControlAI gains traction, the entire business model of major AI labs could face existential regulatory threats. We are moving past the era of voluntary safety commitments. Governments are stepping in because the technology is moving faster than corporate self-regulation. Leahy’s argument forces a uncomfortable question on the industry. Are we building tools we can use, or are we creating entities we can’t control? The answer will define the next decade of tech policy.
