A technique, not a crime, until it is. As of May 2026, the Copyright Alliance had counted over 110 copyright lawsuits against companies building frontier AI models. The accused in these cases are the industry’s biggest names: Meta, Anthropic, OpenAI, and others.
A few of these cases have drawn wide publicity, especially where big tech has been found guilty or has settled. Most centre on the alleged illegal use of copyrighted content to train frontier models. There’s little doubt these models were developed using content that was largely taken without permission. What remains unresolved is the scale of the offence, and what reparations or penalties should follow. Noam Chomsky even suggested replacing the term artificial intelligence with plagiarism software.
So it was a little strange to watch the same companies standing in the dock in the US turn around and accuse Chinese companies of doing the exact same thing they’ve been accused of.
When the Accuser Becomes the Accused
Anthropic named several Chinese companies, alleging illegal distillation, where Chinese firms allegedly trained their own frontier models using responses generated by Anthropic’s Claude. Anthropic published specifics, claiming DeepSeek, Moonshot AI, and MiniMax generated over 16 million exchanges with Claude through roughly 24,000 fraudulent accounts. Earlier, in February 2026, OpenAI sent a memo to US House lawmakers accusing DeepSeek of using distillation to mimic its own products.
For anyone unfamiliar with the term, distillation means collecting a stronger model’s question-and-answer pairs and using them to train a new model that approximates the original’s capabilities, at a fraction of the cost. This is standard practice, and US agencies themselves describe it as a legitimate and useful technique in AI research. The real fight is over doing it at scale, against a rival’s model, in breach of that model’s terms of service.
US companies argue that large-scale distillation lets Chinese firms shorten development time and cost, while American companies continue spending billions of dollars building their own models from scratch. But there’s a subtext here that gets conveniently ignored: these same US companies have spent billions of dollars scraping content from legitimate copyright holders, building models on foundations that are themselves shaky. So how does one alleged offence get treated as unacceptable, while a structurally similar one gets waved away?
Is sauce for the goose not sauce for the gander?
The Fight Goes Official
The US government formally entered this fight on September 8th, when CISA, the NSA, and the FBI jointly released advisory AA26-251A. The advisory accused six China-based AI companies, DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun, and Z.AI, of running industrial-scale distillation campaigns against US frontier models since at least late 2024. According to the advisory, these companies extracted billions of tokens across millions of requests from Claude, GPT, Gemini, and Grok, likely with the backing of the Chinese government.
China responded quickly. Its Commerce Ministry called distillation neutral and widely practised by companies worldwide, including American firms, framing the US accusations as reflecting anxiety and double standards. It also warned of firm countermeasures if the US moves against Chinese AI companies over this issue. China’s Foreign Ministry urged Washington to stop what it called unfounded accusations and smears, crediting China’s AI progress instead to self-reliance.
US Treasury Secretary Scott Bessent added his own voice to the debate, arguing that because Chinese companies are copying US models, they can never get ahead of us.
What This Is Really About: Money
This is where the real issue lies, and it’s less about principle than about economics. Chinese models are inexpensive, often free, while offering comparable computing power to their US counterparts. For an enterprise buyer, nationalistic sentiment tends to lose out to the bottom line. As a result, US companies may face genuine financial losses unless Chinese models are restrained or penalised for illegal distillation.
Given the enormous valuations of US AI companies, and the sheer amount of money they’re pouring into compute, any threat to their expected returns could have serious consequences, not just for individual firms, but for the economics of the entire industry. The ripple effects could extend across the broader US tech sector and economy, where close to 38% of the equity market is tied to tech. If even one major AI company misses two consecutive quarters of revenue forecasts, the hit to its stock price would likely be swift, and potentially severe.
An Unsettled Fight
The tension here is real, even though the legal status of distillation itself remains unresolved. A terms-of-service violation isn’t the same thing as theft, and most of the hard numbers in this dispute come directly from the accusing parties themselves. The Chinese companies named in these allegations have largely stayed composed in their public responses. In the end, the jury is still out, not only on whether Chinese distillation was illegal, but on whether US companies’ own use of copyrighted content to build their models was any different.
