The alarm sounded by AI chief executives about the very products they have raced to market amounts to a remarkable admission of failure and a spectacular failure of governance. Why, then, are global politicians asking the same CEOs to talk about solutions while debating whether their technology has doomed humanity? Once again, we are allowing corporate leaders to dictate the terms of the discussion while asking them what should be done about it. Even European Commission president Ursula von der Leyen, in her recent State of the Union address, invited AI CEOs to discuss “how to pace the frontier” and “support ongoing industry efforts”.

This overture sends the wrong signal at the wrong time. Imagine asking oil companies to design climate policies or car companies to make traffic rules. It is especially peculiar as Europe has a comprehensive AI law, the EU AI Act, with risk mitigation at its heart.

Instead of asking business leaders what society needs going forward, political leaders should be doing a better job of stress-testing their claims and making sure that their businesses and their products contribute to society-wide success. Where were lawmakers when data centres were erected without transparency of ownership, resource demands or the impact on local communities? Where was the oversight when safety concerns were pushed aside to boost competitive advantage? The law should be the anchor of any action by European leaders and should be updated when technological advances merit as much. The AI Act currently includes obligations on transparency, assessing and mitigating systemic risk and foresees oversight of general-purpose models.

These are the very building blocks that many experts are hoping regulation will ensure elsewhere too. Europe should not arrange summits with AI CEOs as a substitute for enforcing the rules it already has. Enforcing existing laws more forcefully allows for more immediate action.

Society requires a far better understanding of what is happening inside AI labs, from experimentation to deployment. That will allow us to make risk assessments, inform policies and pace or even halt new model releases accordingly. Nonbinding moratoria negotiated between the companies that would have to observe them will not get us to a better place.

In fact, earlier calls by AI leaders for a lull in frontier model development led to zero change in their behaviour. We need a comprehensive governance reset led by democratic leaders and institutions to achieve independent scrutiny and real accountability. Dario Amodei of Anthropic has called for a coalition of democracies to govern AI safety together. As the host of leading frontier labs, the US holds the key to such a coalition.

When American companies are accountable at home, the impact ripples around the world. Washington would enjoy its own version of the Brussels effect: American rules would become the default global AI governance standard, the way EU rules on data protection and product safety have shaped norms far from its borders. That effect only materialises, however, if the rules have substance and align in values with those of other democracies.

If they want to see global AI governance then tech companies should be lobbying Washington for binding regulations, a 180-degree turn from the approach being taken. How did we get to a point where AI companies have convinced investors and markets that they will build the future only to warn us that they will end it at the same time? Let us focus instead on strengthening democratic governance of AI. We cannot wait for another incident report warning of disaster.

Pyromaniacs do not make good firefighters.