Most policy debates treat AI as a unique anomaly demanding emergency powers, but Arvind Narayanan & Sayash Kapoor argue that this panic ignores a simpler, more effective path: building societal resilience. They challenge the prevailing narrative that we must restrict innovation to survive, suggesting instead that the real danger lies in granting the executive branch unchecked authority over research and speech.
The Trap of "Abnormal" Technology
The piece opens by dismantling the idea that AI requires a special regulatory regime. While some observers point to emerging cyber and bio-risks as proof that AI is fundamentally different from past technologies, Narayanan & Sayash Kapoor contend that this distinction is often overblown. They write, "Thompson argues that AI seems particularly 'abnormal' because of risks that are emergent and unknown even to AI developers. And this justifies plans by the government to treat AI as an abnormal technology." This framing is compelling because it exposes a logical leap: just because a risk is hard to predict does not mean the only solution is to halt development. The authors suggest that treating AI as an exception rather than a continuation of technological progress risks creating a permanent state of emergency governance.
"Extraordinary interventions impose restrictions on the liberty of actors who are not directly responsible for the harms in question."
The authors define "extraordinary intervention" with surgical precision, noting that such measures often restrict the freedom of builders to punish potential bad actors. This is a crucial distinction. By targeting the companies that create tools rather than the individuals who misuse them, the government risks cutting off beneficial access for the public. Critics might argue that in the face of existential threats, the cost to liberty is a necessary trade-off, but Narayanan & Sayash Kapoor push back, reminding us that "the burden falls not on the malicious actors who cause harm, but on companies that build tools that could, in principle, be misused." This approach mirrors historical struggles where the cure was worse than the disease.
The Illusion of Nonproliferation
A central pillar of their argument is the futility of trying to replicate nuclear-style nonproliferation in the digital realm. Unlike enriched uranium, which requires physical bottlenecks, the core techniques for building AI are well known and easily replicated. Narayanan & Sayash Kapoor write, "AI is different from nuclear weapons. For one, there is no equivalent 'physical' bottleneck. The core techniques for building AI systems are well known." This is a devastating critique of the current policy obsession with export controls and licensing. They point out that even with strict controls, the gap between frontier models and public access is shrinking to mere months, not years.
The authors draw a sharp parallel to the history of cryptography. In the 1990s, the government attempted to restrict encryption software, arguing it would help criminals evade law enforcement. They even investigated a programmer under the Arms Export Control Act for releasing code. Yet, as Narayanan & Sayash Kapoor note, "These restrictions were eventually rolled back... Encryption became the foundation of digital security, enabling e-commerce, online banking, and many other applications." This historical context is vital; it shows that attempts to suppress dual-use technology often backfire, stifling the very defenses needed to secure society. The lesson is clear: trying to wall off technology rarely works, but making society robust enough to withstand it is a proven strategy.
"Resilience distributes defenses across society."
Instead of a single chokepoint, the authors advocate for a distributed defense. They argue that we have successfully managed similar transitions before, such as the rise of automated vulnerability detection tools. These tools, once feared for empowering attackers, became the backbone of modern cybersecurity because defenders had access to them too. "Since defenders had access to the same tools, they became core defensive tools, largely funded by the cyberdefense ecosystem," they explain. This suggests that the offense-defense balance can be managed through investment and adaptation rather than prohibition. A counterargument worth considering is that AI's speed of iteration might outpace our ability to build resilience, but the authors maintain that betting on a brittle nonproliferation regime is a far riskier gamble.
The Slippery Slope of Executive Power
Perhaps the most urgent warning in the piece concerns the long-term impact on democratic governance. If we accept the premise that AI is an "abnormal" threat, we open the door to permanent expansions of state power. Narayanan & Sayash Kapoor ask, "Would they support restrictions on open-weight models? What about requiring approvals for each new model release, or restrictions on the movement of researchers who build frontier AI across countries?" They warn that without clear limits, the demands for government control will escalate as capabilities advance. This is not just about AI; it is about the precedent it sets for all future technologies.
The authors highlight the danger of bypassing normal governance processes. "Extraordinary interventions bypass normal processes of governance, and instead rely on unilateral authority such as emergency declarations or executive orders," they write. This bypasses the democratic accountability that exists to protect liberty. The piece forces us to confront a difficult choice: invest in the slow, unglamorous work of improving resilience, or accept a future where the government decides what research is safe to publish. As they put it, "Nonproliferation is brittle because it relies on a single chokepoint. Resilience distributes defenses across society."
Bottom Line
Narayanan & Sayash Kapoor make a powerful case that the rush to regulate AI as an existential emergency is a strategic error that threatens both innovation and liberty. Their strongest argument is the historical evidence that suppression of dual-use technology fails, while resilience succeeds. The biggest vulnerability in their position is the assumption that society can adapt fast enough to AI's rapid evolution without some form of pre-emptive restriction, but their call for a shift from control to preparation is a necessary corrective to the current panic. Watch for how policymakers balance these competing pressures in the coming months.