Anthropic urged the United States in May 2026 to maintain a 12-24 month lead in advanced AI capabilities over China by 2028 to ensure AI safety and governance [1, 2]. The company described the prospect of China, led by the Chinese Communist Party, taking the lead in frontier AI as a major threat capable of enabling repression at an unprecedented scale [1].
Anthropic warned that China is closing the AI gap through loose controls on chip exports and "distillation attacks," a process of illicitly extracting knowledge from AI models to train competitive versions [2]. In February 2026, Anthropic accused three Chinese AI firms—DeepSeek, MiniMax, and Moonshot AI—of illicitly using Anthropic's Claude model to train their own systems [2].
To counter this, Anthropic advocated strengthening US chip export controls, increasing enforcement budgets, and taking measures to stop distillation attacks by Chinese labs [2]. The US first imposed chip export controls on China in 2022 under the Biden administration, with some partial relaxations occurring in August 2025 under the Trump administration, such as allowing Nvidia sales of H200 chips to China subject to a 25% levy [2].
Competition between US and Chinese AI labs is increasing pressure to release new models faster without adequate safety testing, raising broader safety concerns [2]. Anthropic said, "Our past success means that our present task is largely to avoid squandering our advantage: to decide not to make it easier for the CCP to catch up" [2].
Some experts criticized Anthropic's framing of the US-China AI competition as an arms race. Alvin Wang Graylin, Digital Fellow at Stanford Institute, said the "arms-race framing pushes us in the wrong direction at exactly the wrong moment" [1]. Critics called the framing irresponsible and fearmongering [1].
Anthropic responded that if the US and allies act now, it may still be possible to lock in a 12-24 month AI lead, but the window is closing rapidly [2]. The company urged prompt action to maintain technological and safety advantages for the US in frontier AI development [2].
