The AI regulation trigger has to move. The market will not fix the system on its own

Thesis: two regulatory poles, Washington and Beijing, are saying the same thing this week, though for different reasons. Industry self-regulation is not enough. The difference lies in when the state steps in, not whether it does at all.

Bill Gates appealed to the US Congress in an interview with NBC News to regulate the development of artificial intelligence. "Nobody thinks industry self-regulation alone is enough," he said, asked directly whether Washington has to intervene. His answer was unambiguous: yes, and the requirements have to be mandatory. He admitted this puts some burden on the industry, but in his view it will not meaningfully slow its work. Gates did not deny the risk that AI could threaten humans within a few years. He added a second thread: bad intentions combined with the latest AI tools can now cause damage on a mass scale.

A few thousand kilometres away, in Shanghai, the economist Liu Yuanchun put forward a thesis that sounds like a complement to Gates's statement rather than its opposite. Liu, the rector of Shanghai University of Finance and Economics, argues that AI is not an ordinary technological change but a "supertechnology". Its effect on economic growth, on time and on the way things operate goes beyond the existing notion of technology. The regulation trigger therefore has to move much earlier: the state should plan ahead instead of reacting to the consequences.

What Liu argues from is interesting. He cites an open letter of 13 July 2026, signed by nearly 200 researchers and entrepreneurs, including 16 Nobel laureates. It set out three main points: that this AI differs from previous technologies, that its effects exceed the imagination, and that governments have to build oversight and safeguard mechanisms. He also points out that economists who until recently were sceptical about AI, such as Acemoglu, have changed their minds and now study its impact on employment structure and income stratification. Liu's conclusion is unambiguous: the belief that a "spontaneous order" will settle itself gently has to be abandoned.

A third statement this week leads to the same conclusion, though it comes from a completely different place. Jensen Huang of Nvidia said that if frontier labs cannot keep their systems safe, "we have to shut them down". He pointed out that AI experiments in which runaway agents commit violations create civil and criminal liability that the labs bear. It is telling that this comes from a company selling accelerators, the player that benefits most from fast AI development. Even it accepts that the limit is control, not speed.

The juxtaposition of the three voices is striking. Washington should pass a law, Beijing should move the trigger, and the largest hardware maker allows for labs being switched off. Markets dislike that kind of ambiguity, but politics can no longer pretend not to hear it. What counts today is not whether to regulate, but from when. Every month of delay raises the cost of fixing a system that nobody fully controls. The era in which a manufacturer's voluntary declaration was enough ended the moment models began to act on their own.