"Tighten controls on advanced compute to PRC labs, disrupt their efforts to distill America's best AI models and accelerate democracies' adoption of AI," Anthropic urged.
Anthropic's warning and the forecast to 2028
Anthropic published a report warning that the United States risks losing its frontier artificial intelligence lead to China unless policymakers act to shore up export controls and model defenses. The paper carried a stark prediction in its subtitle: "New Report Warns China Could Reach Frontier AI Near-Parity by 2028." While the firm acknowledged that the U.S. and its allies continue to hold a "significant advantage in frontier AI development," it said that advantage "will narrow quickly" without tighter enforcement of chip export controls and stronger protections for advanced American models.
Chips and hyperscale compute as the defining strategic variable
The report framed advanced semiconductors and hyperscale compute infrastructure as the "defining strategic variable" shaping the balance of power between Washington and Beijing. Anthropic argued that access to powerful compute is not merely an economic input but a strategic asset: powerful frontier AI systems, the firm said, are likely to become deeply intertwined with future cyber operations, military planning, scientific discovery and economic competitiveness over the next several years.
Those stakes help explain why Anthropic urged the White House to tighten controls on advanced compute reaching PRC labs. The warning came amid reporting that the Trump administration may be softening parts of its AI containment posture: CNBC reported that roughly 10 Chinese firms may receive approval to purchase Nvidia H200 chips. Anthropic presented this policy axis — controls on high-performance chips and compute — as central to whether the U.S. can preserve a durable lead.
Talent recruitment and state-backed industrial policy
Anthropic highlighted the speed of Chinese progress, pointing to rapid gains by Chinese AI firms driven by aggressive talent recruitment and state-backed industrial policy. The report cited recent high-profile personnel moves as examples: former Google research executive Wu Yonghui joining ByteDance's AI division, and former OpenAI researcher Yao Shunyu becoming Tencent Holdings' chief AI scientist. Anthropic used these moves to illustrate how human capital, combined with access to compute, accelerates capability development abroad.
Policy signals: summit timing and governance talks
The report was released on the heels of President Donald Trump's summit with Chinese leader Xi Jinping and appeared one day after Treasury Secretary Scott Bessent confirmed that the U.S. and China have begun discussions around AI guardrails and governance frameworks. Anthropic framed those diplomatic and policy conversations against a backdrop of competition pressures that, the firm warned, could incentivize governments and companies to prioritize deployment speed over model security, safety testing and broader governance controls.
What this means for the White House, defense & scientific research sectors, and Chinese AI firms
- The White House: Anthropic explicitly called on the administration to "tighten controls on advanced compute to PRC labs" and to "disrupt their efforts to distill America's best AI models." For the executive branch, the report places export controls and enforcement at the center of retaining a competitive edge.
- Defense and scientific research sectors: Anthropic argued that maintaining a "sustainable democratic lead" will require broader adoption of frontier systems across defense, scientific research and critical infrastructure. These sectors, the firm suggested, must both consume frontier AI capabilities and help secure them against replication.
- Chinese AI firms: The report highlighted that firms in China have accelerated via talent recruitment and state-supported programs; cited moves to ByteDance and Tencent as concrete cases. Anthropic warned that, given compute access and rapid talent flows, near-parity on frontier capabilities is plausible within a defined time horizon.
Anthropic's report layers three linked facts: advanced chips and hyperscale compute matter strategically; talent and industrial policy can accelerate gains; and current signs of shifting U.S. policy — including reported approvals for Nvidia H200 purchases and nascent talks on AI guardrails — intersect with those dynamics. The firm's prescription is concrete and consequential: enforce tighter controls on advanced compute exports, make it harder to replicate U.S. frontier models abroad, and accelerate democratic adoption of AI across public-facing sectors.
The real test will be whether policymakers treat chips and compute as transient trade issues or as levers of long-term strategic posture — and whether enforcement keeps pace with the technical means to replicate advanced models. Anthropic's report leaves that choice plainly on the table.
