ByteDance imposes strict ban on distilling competitor AI models: how Beijing responds to US accusations

Amidst escalating technological rivalry between the US and China, one of the world's largest tech giants — ByteDance — has made an unprecedented decision. The company's founder has imposed a strict ban on its artificial intelligence researchers from using the distillation method on models owned by foreign competitors. This decision is a direct response to the tense geopolitical climate and attempts by US regulators and corporations to accuse Chinese companies of unfair competition.

Ban as a precautionary measure

According to the Pekingnology newsletter, the new policy has been in effect since 2023, but its relevance and strictness have come to the forefront again in 2026. The primary motivation for ByteDance was to avoid new sanctions from Washington. Unlike several other Chinese labs, ByteDance has not yet been officially named among companies using the intellectual property of American giants, such as Anthropic and OpenAI, to train their systems. Maintaining this status is a critical commercial priority for the company, which seeks to preserve access to Western markets and technologies.

What is distillation and why is it dangerous

At the heart of the conflict lies the technology of distillation. This method allows for training a new, more compact and cheaper model (the "student") using a more powerful and resource-intensive model (the "teacher"). This enables the transfer of capabilities of advanced systems with significantly lower computational costs. It is precisely this process that has sparked outrage in Silicon Valley. Previously, the American company Anthropic accused several Chinese firms, including Moonshot, DeepSeek, MiniMax Group, and Alibaba Group, of unauthorized use of their models. OpenAI also raised similar claims, targeting DeepSeek.

Internal debates and competitive pressure

The implementation of the ban did not go smoothly within ByteDance itself. Fierce debates erupted within the company, as refusing to use advanced foreign models for training could slow down the development of their own algorithms. The situation is exacerbated by the fact that ByteDance's best models in some segments have already been surpassed by smaller Chinese labs, which may be less cautious in their methods. Beijing as a whole aims to create more sophisticated AI with fewer computational resources, and currently, American models offer the best price-to-performance ratio. Analysts predict that technology convergence will continue, but the methods of achieving goals will differ.

Unification of Western giants

The situation regarding alleged intellectual property theft has forced OpenAI, Anthropic, and Google competitors to set aside their differences for a common cause. The companies have begun to cooperate to jointly curb the actions of Chinese competitors. This cooperation is aimed at monitoring the market and identifying cases of using their data to train other models. In response, ByteDance, positioning itself as a "white crow" among Chinese developers, is trying to distance itself from accusations by betting on the legitimacy of its development methods.

Contradictory data

There is a discrepancy in assessments of the effectiveness of this policy. On the one hand, ByteDance representatives and their analysts claim that the ban allows the company to maintain its reputation and avoid direct trade restrictions, which is beneficial in the long term. On the other hand, critics within the industry and independent experts point out that refusing distillation puts Chinese models in a losing position compared to Western counterparts, which have access to huge computational resources and do not limit themselves in optimization methods. Furthermore, it is unclear how effectively ByteDance can control its employees and contractors, given that other Chinese companies continue to use controversial methods.