Experts flag China’s low-cost AI strategy as a U.S. risk

Copying capabilities at lower cost

The technique allows a smaller system to learn from the responses and capabilities of a more advanced model. Siskind said that approach can help Chinese developers bypass some of the research, computing and energy costs involved in building a leading model from the beginning.

Gordon Chang, who writes about China, said the United States has more than 5,400 data centers compared with roughly 450 in China and also has more advanced computer chips. Knowledge distillation offers Chinese laboratories a way to work around that hardware disadvantage, Siskind said.

Free models compete with costly U.S. systems

China is also emphasizing free, openly available models while major American developers rely on costly proprietary systems. Siskind said that price difference has led many Fortune 500 companies to consider or use Chinese models for business operations.

She warned that users may be unable to determine whether those models contain Chinese government censorship or hidden access points. Those concerns describe possible risks; the experts did not identify a specific compromised company or model.

Chang compared the strategy to China’s approach to solar panels: lowering costs and expanding availability to gain market share. He argued that widespread adoption could eventually give Chinese developers greater influence over global AI use.

Security questions extend beyond software

Siskind and Chang also warned that widely used AI systems could support cognitive warfare, including attempts to shape people’s beliefs and decisions. The policy questions for Washington include how to assess foreign models, protect companies using them and encourage competitive American alternatives that are less expensive or openly available.