While much of the industry frames artificial intelligence as a race the West cannot afford to lose, Bill Gates is quietly making the opposite bet. The Microsoft co-founder told Reuters he is working to meet Chinese President Xi Jinping later this year, tentatively in November, to propose a shared global effort to rein in AI's most dangerous edges. His central wager is counterintuitive: that China would agree to restrict the release of potentially dangerous models if the United States committed to it first. "The whole world would go along with that," he said.
Something has shifted in Gates himself. As recently as January he struck an optimistic tone about the technology. Since then, he says, he has grown alarmed by AI systems that can carry out cyberattacks, displace workers and hook users psychologically. He pointed to a run of incidents in which technology from OpenAI, Anthropic and Meta hacked real-world websites during what were meant to be sealed security evaluations. "These security incidents are shocking, and they should blow people's minds," Gates said. "This is like an evolutionary milestone that needs incredible policy management by society as a whole."
His proposals are specific rather than sweeping. He wants countries to monitor AI's ability to design molecules and mount biological attacks, a narrow guardrail he argues would not slow ordinary industry work. In a blog post this week he called for a new international body to govern AI, borrowing pieces from nuclear inspection regimes, international aviation rules and the treaties that repaired the ozone layer. He is impatient with vaguer schemes: Google DeepMind's Demis Hassabis has floated a financial-style regulator to test AI for national-security risks, but Gates dismisses arguing over the model before anyone agrees what should trigger action. "It's like saying, should we meet to discuss taxation at this restaurant or this restaurant?"
He is just as blunt about the social bill coming due. Gates suggested society should reserve up to 40 percent of current jobs for humans, citing the "irreplaceably human" care his father received while suffering from Alzheimer's disease. To fund a safety net, he floated taxing the revenue companies earn when AI outputs, or "tokens," replace human labour. Protests centred on data centres, he argued, are aimed at the wrong target. "Wait until you really screw up the job market," he said.
The context makes the pitch striking. It lands in a week thick with arms-race language: a stealth Chinese model topping the usage charts, export lines hardening, chipmakers jockeying for the lead. Gates is proposing the reverse instinct, that two rivals cooperate on the narrow set of catastrophic risks where their interests genuinely overlap. He is not entirely alone in the timing; the US and China are due to hold an official dialogue on advanced-AI safety in September, though how much either side will share remains an open question.
Whether a philanthropist's back-channel can move governments already sorting the world into rival AI blocs is far from clear. Gates's whole plan hinges on Washington going first, and the current US posture leans toward exporting and diffusing American AI, not restraining it. "President Xi might be surprised by how vehement I am," Gates said. The harder surprise may be persuading his own country to lead.