Anthropic published a paper arguing the US can lock in a 12-to-24-month lead over China in artificial intelligence — but only if Washington acts quickly on three specific policy fronts. The company warns that Chinese labs are already using smuggled chips and illicit model copying to narrow the gap, and the window to cement an advantage will close by 2028.
Three policy measures to lock in the lead
The paper lays out three areas where tighter rules could make the difference. First, it calls for cracking down on chip smuggling, restricting foreign access to US data centers, and tightening controls on semiconductor equipment. Second, it says the US should legally clarify that distillation attacks — where a rival copies an AI model by repeatedly querying it — are illegal, and invest in better ways to detect and prevent them. Third, it wants Washington to promote global adoption of trusted US AI hardware and models, creating a self-reinforcing ecosystem.
Anthropic says Chinese labs rely on illicit compute access — chips smuggled in, or data centers located in third countries — and illicit model access through distillation to stay competitive. The paper argues that without these leaks, China’s progress would slow dramatically.
The gap in compute and capabilities
One study cited in the paper estimates that if the US tightened controls on the Chinese Communist Party’s access to US compute, Washington would end up with 11 times as much computing power as Beijing. Another figure, from Stanford’s AI Index, puts the overall US-China AI gap at just 2.7%. That narrow margin underscores how quickly the race could flip.
The paper doesn’t claim the US is guaranteed to stay ahead. It says the advantage depends on enforcement and execution. If current leaks continue, the Chinese side could close the distance well before the end of the decade.
The narrow window and the stakes
Anthropic stresses that the window to lock in a lead is narrow. “The window to lock in a lead by 2028 is narrow and will not remain open for long,” the paper states. It warns that AI capabilities could soon be used to repress citizens on an unprecedented scale and could reshape the global balance of power. That’s not a distant scenario — the company says it’s a near-term risk if policy doesn’t keep pace with the technology.
The paper is a direct appeal to US policymakers. It’s not just about trade policy or export controls; it’s about national security and the future of democratic governance. Anthropic is betting that a clear, specific road map will push the conversation beyond broad goals toward concrete action.
Whether that action arrives in time is the open question.
