US Agencies Warn of Industrial-Scale AI 'Capability Theft' by Chinese Firms
CISA, FBI, and NSA report that China-based companies are extracting billions of tokens from U.S. frontier models to shortcut R&D costs.
The U.S. government has issued a stark warning that China-based artificial intelligence companies are systematically stripping proprietary capabilities from American AI models. In a joint cybersecurity advisory released September 8, 2026, CISA, the FBI, and the NSA detailed industrial-scale "knowledge distillation" campaigns designed to accelerate China's AI development at the expense of U.S. intellectual property.
According to the advisory, these campaigns involved the extraction of billions of tokens from frontier AI models, including those developed by OpenAI, Anthropic, Google, and xAI. To achieve this, attackers executed millions of requests, deliberately bypassing terms of use by routing traffic through a complex web of pathways. These included native APIs, cloud providers, third-party aggregators, relays, and vendor account pools.
The Mechanics of Distillation
Knowledge distillation is a recognized AI research technique where a smaller "student" model is trained to mimic the outputs and behavior of a larger, more complex "teacher" model. While legitimate in a research context, the U.S. government asserts that Chinese firms have weaponized the process. By treating U.S. frontier models as the teachers, these firms can effectively "shortcut" the massive research and development investments typically required to build high-performing models from scratch.
A Shift to Capability Theft
This activity marks a critical evolution in cyber espionage, shifting from the traditional theft of static data to what is described as "capability theft." By extracting the underlying logic and performance of U.S. models, adversaries can achieve technical parity without the corresponding investment in massive compute clusters or curated datasets.
U.S. officials warn that this allows China-based companies to see "significantly shorter AI development timelines and reduced financial expenditures in training a frontier model," according to the government advisory. This erosion of the U.S. competitive advantage poses a direct risk to both the commercial AI market and broader national security interests.
Industry Implications
As AI companies continue to deploy powerful models via APIs, the vulnerability to automated extraction grows. The scale of these campaigns suggests a highly organized effort to bridge the gap between Chinese and American AI capabilities. While the advisory highlights the systemic nature of the threat, the industry now faces the challenge of distinguishing legitimate API usage from industrial-scale distillation attempts.
Moving forward, the focus will likely shift toward more aggressive monitoring of request patterns and the implementation of stricter safeguards to prevent the wholesale cloning of model intelligence. This systemic threat underscores the fragility of the current API-based delivery model, where the very accessibility that drives innovation also provides the primary vector for intellectual property theft on a global scale.