U.S. Accuses Chinese Firms of Systematic Theft of Advanced AI Models

Federal authorities have charged a group of Chinese artificial‑intelligence companies with conducting a coordinated, illegal effort to extract and replicate the capabilities of cutting‑edge U.S. AI systems. The allegation, outlined in a recent joint advisory, marks a rare public naming of foreign actors in the rapidly evolving field of generative AI.

The advisory, issued jointly by the National Security Agency and other cybersecurity agencies, describes the activity as a "systematic" distillation of U.S. frontier models. According to the notice, the Chinese entities employed techniques that reverse‑engineer large language models, capturing their performance characteristics without authorization.

Model distillation involves training a smaller, more efficient version of a larger AI system by feeding it the original model's outputs. While the practice can be legitimate when the source model is openly licensed, the advisory claims the Chinese firms used proprietary U.S. models—such as those powering leading chatbots—as raw material, violating intellectual‑property and export‑control rules.

The accusation arrives amid a broader backdrop of heightened U.S.-China competition over high‑technology dominance. Over the past decade, Washington has repeatedly warned that illicit technology transfer erodes national security and economic advantage. Prior investigations have targeted semiconductor designs and advanced manufacturing equipment; AI now represents the latest frontier in that contest.

Officials say the next steps could include targeted sanctions, restrictions on technology transfers, and potential criminal prosecutions. Industry analysts note that the move may prompt tighter export controls on AI research tools and increase scrutiny of cross‑border collaborations. The advisory also urges U.S. companies to bolster safeguards against unauthorized model extraction, signaling that the government intends to treat AI theft with the same vigor as other forms of intellectual‑property espionage.