ByteDance founder Zhang Yiming reportedly forbade the company's artificial intelligence researchers from distilling rival models in 2023, according to The Information, years before avoiding American retaliation became a commercial priority. The state-backed Chinese outlet The Paper separately reported on 6 August 2026 that Zhang told a recent internal meeting that AI development requires "long-termism and delayed gratification, rather than using others' output to achieve short-term leaderboard rankings", citing sources within the company.
The dating matters because it predates the period when avoiding US sanctions became a priority, as the Pekingnology newsletter noted. A policy written in 2023 cannot have been designed to dodge sanctions that were not threatened until 2026, which suggests either unusual foresight or a genuine research position.
The company that was not named
ByteDance is conspicuously absent from Anthropic's accusations. The US laboratory alleged that DeepSeek, Moonshot AI and MiniMax used roughly 24,000 fraudulent accounts to generate more than 16 million exchanges with Claude in what it described as 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. None of these claims has been proven in court or independently verified. They remain allegations.
Those accusations are what triggered Washington's sanction threats. Treasury Secretary Scott Bessent warned in July 2026 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 2026. Beijing has accused Washington of "AI hegemonism" and threatened countermeasures.
What abstaining reportedly cost
The 2023 guidelines reportedly sparked intense debate inside ByteDance, whose best models have been surpassed by smaller Chinese rivals. Its Doubao-Seed-Code model and Trae coding tool have underperformed against Zhipu's GLM 5 and Moonshot's K2. Moonshot, a fraction of ByteDance's size, 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.
The commercial picture is less bleak than the benchmarks suggest. Doubao reached 345 million monthly active users by March 2026, well ahead of Qwen at 166 million and DeepSeek at 127 million, and ranked second globally by app users behind ChatGPT.
Enforcement and exposure
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.
Proving distillation is the harder problem. The White House policy memo accusing China of industrial-scale distillation committed to intelligence sharing with US laboratories, but detection relies largely on usage patterns rather than anything visible in a finished model. 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 compute constraint
Beijing is pressing its laboratories 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 2026 requiring export licences for model weights and training pipelines leaving the country.
The Economist argued in August 2026 that American models still offer better value but that convergence is likely to continue. If that holds, the strategic question is whether Chinese laboratories can sustain the pace without borrowing from US systems. ByteDance has effectively volunteered to answer it.
What is established and what is merely claimed
Established: ByteDance does not appear among the laboratories Anthropic has accused of distillation. The Paper reported on 6 August 2026 that Zhang Yiming instructed staff to avoid distilling rival systems, and The Information reported that an internal policy dates to 2023. Washington's policy moves, the April 2026 White House reclassification and Bessent's July 2026 sanctions warning, are on the record, as is Beijing's "AI hegemonism" response.
Merely claimed: That ByteDance's policy genuinely dates to 2023 rests on anonymous sourcing and has not been independently confirmed. Anthropic's account-volume figures against DeepSeek, Moonshot, MiniMax and Alibaba remain allegations, as does the broader claim of coordinated extraction. Whether abstention proves to be principle, foresight or an expensive mistake depends on results nobody can yet measure.