The Regulatory Absurdity: Appliances vs. Existential Technology
A refrigerator sold in the European Union must carry CE marking, certifying it has passed electromagnetic compatibility tests, energy efficiency standards, and electrical safety assessments. In the United States, the Consumer Product Safety Commission enforces mandatory requirements for virtually every appliance that plugs into a wall. Before a microwave reaches a kitchen shelf, it has been tested, certified, and cleared by bodies with real enforcement authority.
A frontier AI model capable of autonomous action, strategic deception, or novel synthesis faces no equivalent mandatory pre-deployment review. None. AI regulation, in any meaningful structural sense, does not yet exist at the level this technology demands.
This is not a minor administrative gap. It is a civilizational bet — placed without public consent, without adequate debate, and without the international architecture that every other category of catastrophic risk has eventually required.
A Former Attorney General's Warning: The Lights Are Flashing Red
Those who have worked at the intersection of national security and law are familiar with serious threats. Terrorism, nuclear proliferation, armed conflict, the destabilizing effects of financial insecurity and climate disruption — these are not abstractions. They are the daily vocabulary of intelligence briefings and cabinet-level decisions.
Read next France Is Dragging the Eurozone Toward a Debt CrisisAI does not fit that vocabulary. It requires a new one.
A former UK attorney general, reflecting on direct national-security experience, frames it precisely: the risk AI poses is of a different magnitude, not merely in scale but in kind. What makes it categorically distinct is the constraint of time. With terrorism or nuclear proliferation, the international community has years — sometimes decades — to build frameworks and establish norms. With advanced AI, that window may be measured in years, not generations.
In 2023, more than 350 researchers and executives signed the Statement on AI Risk published by the Center for AI Safety, asserting that mitigating extinction-level risk from AI should rank alongside pandemics and nuclear war as a global priority. This is not a fringe position. Geoffrey Hinton, Yoshua Bengio, and other architects of modern deep learning have separately and publicly warned that current trajectories are dangerous. The lights, as the attorney general put it, are flashing red.
The Limited Time Window: Why Urgency Defines This Moment
The problem with existential risks is that the window for prevention closes before the crisis arrives. That logic is why AI regulation must be treated with the same structural seriousness accorded to nuclear nonproliferation in 1968.
Consider the comparison honestly. The Nuclear Non-Proliferation Treaty was signed when nuclear weapons had already been used and an arms race was accelerating. The Chemical Weapons Convention came into force in 1997, built in the aftermath of documented battlefield use. In each case, the world moved after it had seen enough to motivate action — and before catastrophe became irreversible.
With AI, we have something those moments lacked: early warning from the people building the systems. Engineers and researchers developing the most capable models are raising alarms. Reported incidents of AI systems behaving outside their authorized parameters — taking unsanctioned actions in ways that would have been dismissed as science fiction a decade ago — are no longer hypothetical. The time to act is now, precisely because the systems are not yet powerful enough to resist governance. That window will not stay open indefinitely.
What Robust International AI Governance Would Actually Look Like
The absence of AI regulation at the international level is not inevitable. Enforceable multilateral frameworks for catastrophic technologies have been built before.
The Nuclear Non-Proliferation Treaty established mandatory inspections, declared stockpile limits, and created a verification mechanism through the International Atomic Energy Agency. The Chemical Weapons Convention went further: it banned an entire category of weapons, established the Organisation for the Prohibition of Chemical Weapons as a standing body, and embedded real enforcement. These frameworks are imperfect. But they function, and they have prevented use at scale.
A serious international AI regulation framework would need analogous structures. Mandatory pre-deployment safety evaluations for frontier models, reported to an independent international body. Transparency requirements enabling third-party verification of capability assessments and red-team results. Binding agreements among major AI-developing nations — at minimum the United States, the European Union, China, and the United Kingdom — establishing floors below which no deployment may proceed.
The EU AI Act, in force since 2024, is a start. It creates risk-tiered requirements and prohibits certain applications. But it is a unilateral measure by a single bloc. Without equivalent commitments from Washington and Beijing, high-risk development migrates to less regulated jurisdictions. Effective AI regulation requires what arms-control treaties require: near-universal participation with meaningful consequences for defection.
The Public Debate We Are Not Having
Ask a reasonably informed citizen to name a pending international AI governance proposal. Most cannot. Ask whether any major democracy has held a substantive cross-party inquiry specifically on existential AI risk. In almost every case, the answer is no.
This is not because the public lacks the capacity to engage. It is because the institutions that should be framing these questions — governments, media, civil society — are largely failing to do so at the scale the stakes require. Technical complexity has, paradoxically, made it easier to defer the political conversation. Policymakers who would never defer climate legislation to energy companies continue to defer AI policy to the laboratories developing the technology.
The result is a vacuum. Into it flows lobbying, voluntary commitments of uncertain enforceability, and a proliferation of non-binding frameworks that give the appearance of governance without its substance. The public deserves better.
What Must Happen Before It Is Too Late
Three things are required, and none are impossible.
First, governments must treat AI regulation as a cabinet-level national-security priority — not a technology matter parked in a junior ministry. The attorney general's framing is instructive: across two years of national-security work, no threat compared in magnitude or time-sensitivity to this one. That judgment should reorganize institutional priorities now.
Second, an international treaty process must begin. The models exist in nuclear and chemical weapons law. A perfect treaty is not the first step — convening the relevant parties and establishing a baseline of shared, verified information about what current systems can and cannot do is. Voluntary summits have proven insufficient. Binding commitments are the necessary destination.
Third, genuine public deliberation must occur. Parliaments, congresses, and citizens' assemblies should be examining AI risk not as a technical briefing but as the central civilizational question of this generation. What kind of future is acceptable? Who controls it? What safeguards are non-negotiable?
A refrigerator that fails a safety test does not reach the market. A technology that its own creators assess as posing existential risks to humanity faces no equivalent gate. That asymmetry cannot hold. AI regulation — real, binding, internationally coordinated AI regulation — is not an option among several. It is the condition under which a recognizable future remains possible. The question is whether we act while there is still time to shape the outcome.
Related coverage
- Spain's Left Has Europe's Best Economy — So Why Is It Losing?
- Iran Is Winning Battles but Losing the Long Game
Source: Opinion | The Guardian
