The US-China AI distillation dispute widened after the United States accused six Chinese artificial-intelligence companies of using large-scale model distillation to copy capabilities from American AI systems.
The allegations were made in a September 8 statement from US law-enforcement and intelligence officials, according to Reuters. Companies named in the reporting included DeepSeek, Moonshot AI and Alibaba. The officials said the activity likely occurred with awareness of the Chinese government, but they did not present a public court finding establishing that claim.
What model distillation means
Model distillation is a common machine-learning technique in which a smaller model learns from the outputs of a larger, more capable model. It can reduce the cost and computing power required to build a useful system.
The technique itself is not automatically improper. The legal and commercial questions depend on how training data or model outputs were obtained, the applicable terms of service, intellectual-property rules and whether access controls were bypassed. The US allegation focuses on the claimed scale, targeting and purpose of the activity.
Why technical evidence will matter
Similar behaviour in two AI models does not by itself prove that one was distilled from the other. A persuasive technical case would need evidence such as access records, repeated output patterns, account links or documented attempts to evade usage limits. Independent experts would also need enough information to evaluate whether the observed similarities could result from common training data, standard benchmarks or convergent model design.
That distinction matters because distillation is used throughout the industry for legitimate research and product development. The dispute turns on the alleged method of access and the claimed intent, not on the existence of the technique alone.
Which American systems were allegedly targeted
US officials said variants of models associated with Anthropic, OpenAI, Google’s parent Alphabet and SpaceX were targeted. They alleged that the process helped Chinese developers train products more quickly and at a lower research cost.
The statement also linked the alleged activity to potential military and cyber capabilities. That is a national-security assessment made by US officials, not an independently verified conclusion presented in the public reporting.
China’s response and the diplomatic context
The Chinese embassy in Washington did not immediately respond to Reuters’ request for comment. A complete assessment will require responses from the named companies and Chinese authorities, as well as technical evidence showing how the alleged access occurred.
The claims arrive before a planned visit by Chinese President Xi Jinping to the United States later in September. Washington and Beijing are also expected to discuss AI safety risks in a separate dialogue, making questions about model access, security controls and intellectual property part of a wider diplomatic agenda.
What the dispute could change
American AI developers may tighten rate limits, identity checks, monitoring and restrictions on automated access. Governments could also consider new export controls or reporting requirements. Such measures may protect proprietary systems, but they can also increase costs for researchers and legitimate developers.
Companies will have to balance security against access. Stronger verification can deter organised scraping, yet broad restrictions could disadvantage universities, start-ups and users in countries with limited computing resources. Clear rules, consistent enforcement and an appeal path would reduce the risk that commercial controls are applied unpredictably.
For the global AI sector, the central issue is whether technical safeguards and existing laws can distinguish ordinary competitive learning from prohibited extraction. Public evidence, clear company responses and any formal legal action will matter more than political labels alone.
The Press of Asia will update this report if the named companies or Chinese authorities issue a substantive response.
Reporting basis: Reuters reporting published by The Straits Times on September 9, 2026. Featured image is an AI-generated editorial illustration.
The Tech Desk covers technology, artificial intelligence and cybersecurity. Stories are checked against at least two credible sources before publication.
