Zhang Yiming reportedly told researchers not to train on competitors’ outputs years before it became politically useful, and accepted falling behind Chinese rivals as the price.
Credit: ByteDance
ByteDance founder Zhang Yiming reportedly banned researchers from distilling rival AI models in 2023, years before US sanction threats made abstention politically useful. ByteDance is the one major Chinese lab Anthropic has not accused of distillation, and its models have been surpassed by several smaller Chinese rivals that were named.TL;DR
ByteDance founder Zhang Yiming reportedly forbade the company’s AI researchers from distilling rival models, according to The Information, a policy that dates to 2023. The timing matters because it predates the period when avoiding US retaliation became a commercial priority, as the Pekingnology newsletter noted.
Distillation involves feeding queries to a powerful model, collecting the outputs, and using them to train a cheaper system that approximates the original. It has become the central flashpoint in the US-China AI dispute.
ByteDance is conspicuously absent from Anthropic’s accusations. The US lab alleged that DeepSeek, Moonshot AI, and MiniMax used roughly 24,000 fraudulent accounts to generate more than 16 million exchanges with Claude in a coordinated extraction campaign.
Alibaba’s Qwen operation was accused of the largest known effort, some 28.8 million exchanges using nearly 25,000 fraudulent accounts. Those charges are what triggered Washington’s sanction threats.
Treasury Secretary Scott Bessent warned in July that Chinese firms could face financial sanctions and Entity List blacklisting over what he characterised as model theft. The White House had already reclassified covert distillation by adversarial states as a national security threat in April.
Declaring distillation illegitimate is easier than proving it happened. The White House policy memo accusing China of industrial-scale distillation committed to intelligence sharing with US labs, but detection relies largely on usage patterns rather than anything visible in the finished model.
The ambiguity runs deep. OpenAI’s Greg Brockman called Moonshot’s Kimi K3 “pretty good” while declining to confirm whether it involved distillation, which is roughly where the evidence sits for most of these claims.
The guidelines reportedly sparked intense debate inside ByteDance, whose best models have been surpassed by smaller Chinese labs. Its Doubao-Seed-Code model and Trae coding tool have underperformed against Zhipu’s GLM 5 and Moonshot’s K2.
Moonshot went on to release Kimi K3, a 2.8-trillion-parameter open-weight model that developers ranked above GPT-5.6 Sol and Fable 5 on a blind coding leaderboard. Moonshot is a fraction of ByteDance’s size.
The commercial picture is less bleak than the benchmarks suggest. Doubao reached 345 million monthly active users by March, well ahead of Qwen at 166 million and DeepSeek at 127 million, and ranked second globally by app users behind ChatGPT.
No Chinese technology company has more to lose from American regulatory attention. TikTok has spent years under US political pressure, which makes ByteDance uniquely exposed to any accusation of intellectual property theft.
That exposure explains the caution, but not the chronology. A 2023 policy cannot have been designed to avoid sanctions that were not threatened until 2026, which suggests either unusual foresight or a genuine research position.
Beijing is pressing its labs to build better models with less compute, a demand shaped by export controls rather than preference. Distillation is an obvious shortcut under those conditions, which is what makes refusing it costly.
China is also tightening its own grip on model access. Beijing has weighed curbing overseas access to its top AI models, and issued guidance in July requiring export licences for model weights and training pipelines leaving the country.
The Economist argued this month that US models still offer better value but that convergence is likely to continue. If that holds, the strategic question is whether Chinese labs can sustain the pace without borrowing from American systems.
ByteDance has effectively volunteered to answer it. Whether abstention proves to be principle, foresight, or an expensive mistake depends on results nobody can yet measure.
Get the most important tech news in your inbox each week.
