U.S. Agencies Charge Chinese AI Firms with Large-Scale Model Replication

Federal intelligence and cybersecurity agencies have publicly accused several Chinese artificial‑intelligence companies of running systematic operations to copy cutting‑edge U.S. AI models, a practice officials describe as industrial‑scale "distilling."

The National Security Agency, the Cybersecurity and Infrastructure Security Agency and the FBI released a joint statement naming DeepSeek, Moonshot and other firms as participants in what they call a coordinated effort to extract the capabilities of frontier American models and repackage them for the Chinese market.

According to the agencies, the companies employ a technique known as model distillation, which involves training smaller, more efficient models on the outputs of larger, proprietary systems. By feeding massive amounts of data generated by U.S. models into their own training pipelines, the Chinese firms can reproduce comparable performance without direct access to the original source code or training datasets.

The allegations arrive amid heightened U.S. focus on protecting intellectual property and national‑security interests in the rapidly evolving AI sector. Since 2020, American startups and established tech giants have poured billions into developing large language models and multimodal systems, sparking a global race for leadership in generative AI. Officials say that unauthorized replication undermines that investment and could give adversarial actors a shortcut to advanced capabilities.

Chinese AI companies have long asserted that their products are homegrown, emphasizing rapid domestic talent development and independent research. However, the U.S. statements suggest a pattern of reverse engineering that goes beyond legitimate benchmarking, pointing to “industrial‑scale” operations that leverage the sheer computing power and data access of the United States to accelerate development abroad.

Policy experts warn that the accusations could trigger a series of diplomatic and regulatory responses, ranging from tighter export controls on AI hardware and software to coordinated sanctions against the identified firms. Lawmakers have already begun discussing legislation aimed at bolstering supply‑chain security for high‑performance computing resources that underpin large‑scale model training.

The agencies indicated that investigations are ongoing and that additional enforcement actions may follow. While the public statement provides limited technical detail, it underscores a broader concern that AI technology, once seen as a purely commercial frontier, is now a strategic asset subject to the same espionage and theft concerns that have traditionally surrounded defense‑related innovations.