U.S. Accuses Six Chinese AI Firms of Industrial-Scale Model Copying

Federal officials accuse six companies of targeted model copying

WASHINGTON — United States intelligence and law-enforcement officials accused six Chinese artificial-intelligence companies Tuesday of using outputs from leading American systems to accelerate development of their own models, according to Reuters. The government described the alleged campaign as aggressive, malicious and targeted distillation conducted at industrial scale.

Reuters says the statement identifies DeepSeek, Moonshot AI and Alibaba among the six accused companies. It alleges that variants of models from Anthropic, OpenAI, Google and SpaceX were targeted. The reviewed Reuters report did not name the other three Chinese companies, so this article will not guess their identities.

Officials assessed that the activity occurred likely with Chinese government awareness, Reuters reported. That phrase is an intelligence assessment, not proof of direct government command, funding or participation, and the public statement as reported does not disclose the evidence behind that level of confidence.

Distillation is a normal technique; authorization is the dividing line

Model distillation is a legitimate machine-learning technique in which a smaller system learns from the outputs or behavior of a more capable model. Developers use it lawfully to reduce cost, improve speed and transfer capabilities when they own the systems, have permission or operate within applicable terms.

The accusation is not that distillation itself is inherently theft. Officials and American companies allege that Chinese firms obtained large volumes of model outputs through deceptive accounts, access restrictions or other methods that violated provider rules, then used those outputs to improve competing systems.

Whether particular conduct violates a contract, trade-secret law, export controls or criminal law depends on the evidence, the information obtained, the method of access and the legal authority asserted. No court ruling, criminal indictment or publicly released sanctions order establishing liability for the six companies accompanied the Reuters account reviewed Tuesday.

A prior White House memorandum made the broader charge in April

The new statement follows an April 23 national security memorandum in which the White House said foreign entities, principally those based in China, were conducting deliberate campaigns to distill United States frontier AI systems at industrial scale. The memorandum directed federal agencies to strengthen protection of American AI technology and infrastructure.

That official document confirms that the administration has treated model distillation as a national-security issue for months. It does not independently prove the newly reported allegation against each of the six companies or establish that every interaction with an American model was unauthorized.

The government's stated concern is that copied capabilities could accelerate systems used for military, intelligence or cyber operations. That is the government's risk assessment. The reviewed public material does not demonstrate that a particular distilled model produced a specific military or cyberattack.

Anthropic previously described millions of allegedly abusive exchanges

Anthropic said in February that it had detected what it called industrial-scale distillation campaigns associated with DeepSeek, Moonshot AI and MiniMax. The company alleged more than 16 million exchanges through approximately 24,000 fraudulent accounts.

Anthropic said the operators used intermediaries and other measures to conceal their identities and bypass access controls. Its report provides technical and operational detail, but it remains the account of an interested company enforcing its own terms—not a neutral adjudication of every allegation.

The overlap between Anthropic's named companies and the government statement adds context, but it should not be used to merge distinct claims. The company report addresses activity Anthropic says it observed on its systems; the newer government accusation reaches multiple American providers and six Chinese firms.

The accused companies and Beijing deserve a documented opportunity to respond

The Chinese Embassy in Washington had not immediately responded to Reuters' request for comment when the report was published. This article also found no reviewed public response from every named company addressing the new six-firm accusation.

Silence at the first report is not an admission. A complete public record should include the statement itself, the identities of all six companies, the evidence connecting each company to particular accounts or transactions, the legal rules allegedly violated and each accused party's response.

Government officials can protect sensitive intelligence sources while still disclosing enough methodology, dates and corroborating material for the public to evaluate a sweeping accusation. Without that record, the claim is consequential but remains a claim.

What happens next could affect diplomacy, access controls and enforcement

The accusation arrives as Washington and Beijing discuss possible AI-safety talks and before an expected meeting between President Donald Trump and Chinese President Xi Jinping, Reuters reported. Those diplomatic plans remain separate from proof of the alleged activity, but the dispute could shape the agenda and tone of any talks.

The next material developments would be a complete official assessment, criminal or civil filings, export-control action, platform account restrictions, public technical indicators or company responses backed by evidence. Each would need to be evaluated on its own terms.

For now, the confirmed event is that United States officials issued a new accusation against six Chinese AI firms. The alleged industrial-scale copying, Chinese government awareness and potential military or cyber consequences remain attributed assessments—not established judicial facts.

About the image

The accompanying image is an authentic January 30, 2019 White House photograph by Andrea Hanks of United States and Chinese delegations meeting for trade talks. It provides historical bilateral context and does not depict artificial-intelligence systems, any of the six accused companies, alleged model distillation or Tuesday's government statement.

As an official White House photograph, the image is in the public domain in the United States and is available through Wikimedia Commons. Its use does not imply endorsement by the photographer, the White House, any person shown or Wikimedia Commons.