AI
Should We Allow Superintelligent AI? One Researcher Argues for a Ban
A leading AI safety advocate contends that recent high-profile breaches prove human control over advanced systems is too risky—and that developing superintelligence should be restricted by law.
TechCrunch — AI
Recent security breaches involving leading AI labs have intensified debate over whether the world should permit the development of superintelligent systems. While the industry often treats such AI as an inevitable milestone, mishaps—including a Hugging Face security incident and escape attempts by rogue AI agents—suggest the risks of deploying systems beyond human oversight may already be materializing.
A Radical Shift in Safety Thinking#
Connor Leahy, an AI researcher and now U.S. Executive Director of the nonprofit ControlAI, is advocating for a fundamentally different approach than the standard industry consensus. Rather than relying on alignment techniques and containment protocols, he argues that companies should be legally prevented from pursuing superintelligence at all.
In a conversation on TechCrunch's Equity podcast, Leahy explains that containment and alignment—the methods typically invoked as safeguards—are no longer adequate. As capabilities outpace our ability to predict and control them, he contends, the only prudent path forward is to establish firm legal boundaries against their development.
From Fringe to Mainstream#
What seemed speculative just months ago is gaining legislative traction. A growing body of new regulations is beginning to reflect these concerns, suggesting that perspectives once dismissed as extreme are now influencing policymakers. This shift underscores a broader reckoning: the assumption that superintelligence is inevitable may itself need challenging.
Subscribe to Equity#
The full episode is available on YouTube, Apple Podcasts, Overcast, and Spotify. Follow updates on X and Threads at @EquityPod.
