Leading US AI developer Anthropic has leveled serious accusations against Chinese technology companies. At the Aspen Security Forum, Anthropic's head of national security policy, Tarun Chhabra, stated that Chinese lab Zhipu AI trained its latest model, GLM-5.2, on responses from Claude and GPT. This constitutes a direct violation of intellectual property rights, threatening the competitive advantage of the United States.

According to expert estimates, the current gap between US and Chinese advanced AI models is six to nine months in favor of the US. However, without unauthorized distillation—a method where a less powerful model learns from the responses of a larger one—this gap could have reached as much as a year and a half. Chhabra called Beijing's behavior "hostile," emphasizing that competitors are extracting the most valuable intellectual property from others' models.

Distillation is not an isolated incident but a systemic practice. Almost all leading Chinese labs use Claude, GPT, and other US models to train their systems. Anthropic is actively blocking accounts involved in this, up to several million per week. Special attention is being paid to Zhipu AI and DeepSeek, which, according to the company, continue a "hostile campaign" to extract data.

Independent Research Confirms Allegations

Research firm Graphistry has published data supporting these allegations. According to Cohen's kappa coefficient, GLM-5.2's responses matched GPT-5.5 at 0.8 and Opus 4.8 at 0.76. For comparison, the coefficient between GPT-5.5 and Opus 4.8 themselves was only 0.63. Zhipu AI has neither confirmed nor denied these assessments, leaving room for speculation but not changing the essence: the evidence of technology theft is becoming increasingly compelling.

Geopolitical Context and Countermeasures

This incident is just part of a broader technological confrontation between the US and China. In July, Chinese authorities called Anthropic's Claude Code a "serious threat" to Chinese users, after which Alibaba banned its employees from using the tool. In parallel, Washington is tightening AI policy: the Gold Eagle platform has been launched, where leading US companies share cybersecurity vulnerability data, and Treasury Secretary Scott Bessent stated that regulators will continue to use advanced AI models to "stay ahead of adversaries."

Recall that in February, Anthropic already accused DeepSeek, Moonshot AI, and MiniMax of distilling Claude through thousands of fake accounts, and later called the Alibaba-linked campaign the largest known attack of its kind. These events underscore that the battle for AI leadership has entered an active phase, and unauthorized use of technology is becoming a key tool in this war.

Expert Opinion: Distillation is not just a technical nuance but a direct threat to innovation. Systematic copying of Claude and GPT responses allows Chinese labs to close the gap without their own investments in fundamental research. If the US does not strengthen the protection of its intellectual property, we risk losing not only a temporary advantage but also the very incentive to create breakthrough technologies. The market must understand: such practices destabilize the entire AI ecosystem.