Palantir CTO warns Chinese AI firms using distillation attacks on U.S. tech

Palantir Technologies, a U.S. data software company serving government and commercial customers, has raised alarms about what it sees as a systematic threat to American AI dominance. Chief Technology Officer Shyam Sankar warned that Chinese developers are building advanced artificial intelligence models through distillation attacks—a technique in which developers train a lower-cost model by learning from outputs generated by another AI system. Sankar characterized the practice as both an economic threat to the United States and evidence that major American AI laboratories have a direct financial incentive to protect their intellectual property more aggressively.

The warning comes in the context of mounting evidence that the technique is already in use. Last month, Anthropic, the U.S.-based developer of the Claude AI model, accused Alibaba Group Holding, a Chinese technology company, of conducting a large-scale distillation campaign through thousands of fraudulent accounts. That accusation has been accompanied by broader concerns from American AI companies that Chinese competitors—specifically the AI startups DeepSeek and MiniMax—may have employed similar methods to create rival chatbots at significantly lower cost than building original models would require. For investors, the dispute highlights an emerging risk: as U.S. AI developers spend heavily on frontier models, overseas competitors may be able to reproduce comparable capabilities more efficiently by circumventing the original development process.

Sankar also identified a second, potentially larger economic risk: growing U.S. opposition to AI data centers. On Tuesday, New York Governor Kathy Hochul ordered a moratorium on new large data centers. Sankar argued that retreating from AI development infrastructure could carry major long-term consequences for the United States. In a separate comment, he acknowledged that France's recent decision to replace Palantir's data tools at its domestic intelligence agency with a local alternative reflected a broader European push for greater technological sovereignty, though he criticized what he described as unfair efforts to weaken Palantir's position outside the U.S. These developments position Palantir at the intersection of several powerful trends: rising AI adoption, tighter technology competition between the U.S. and China, and increasing government preference for domestically developed systems.

Palantir's warning reflects a widening concern among U.S. AI companies about unauthorized knowledge transfer to Chinese competitors. Anthropic's recent accusation against Alibaba—conducted through thousands of fraudulent accounts—demonstrates that distillation is not merely a theoretical threat but an active practice. The involvement of both established players like Alibaba and newer competitors such as DeepSeek and MiniMax suggests the technique has become a preferred route for Chinese developers to close the AI capability gap without bearing the full R&D cost that American labs incur.

For U.S. investors and policymakers, this dispute surfaces a broader tension: American AI companies have invested heavily in frontier models under the assumption that their intellectual property would generate competitive returns, yet distillation allows rivals to capture much of that value at lower cost. Sankar's call for more aggressive intellectual-property protection implies that current safeguards—whether contractual, technical, or legal—may be insufficient. The concern is not hypothetical; if Chinese startups can match U.S. model performance at a fraction of the cost, it reshapes the economics of the entire AI market, potentially eroding margins and innovation incentives for American firms.

Sankar's broader warning about New York's data-center moratorium adds another dimension to the competitive landscape. He framed retreating from AI infrastructure development as a long-term economic risk for the United States. Together, these comments suggest that Palantir and other stakeholders view U.S. AI leadership as vulnerable on multiple fronts: intellectual-property leakage, infrastructure constraints, and geopolitical preference for domestic technology sovereignty (as evidenced by France's recent decision to replace Palantir's tools at its intelligence agency).

No comments yet. Be the first to share your thoughts!