China’s open-weight AI models to be spared US tests, sources say

Washington – The White House has told top US artificial intelligence companies that open-weight models being developed by their Chinese rivals will not be subject to government testing under the Trump administration’s new AI safety framework, according to people familiar with the matter.

The decision to spare open-weight models, which allow users to download and customise the technology, was delivered in a closed-door White House meeting on Aug 4 between United States officials and industry attendees, according to the people. OpenAI, Anthropic and Alphabet’s Google were among the Silicon Valley firms expected to join.

The safety framework, which has not yet been released and may remain private, was called for under an executive order signed in June by US President Donald Trump aimed at addressing AI security issues. Trump’s order outlined a voluntary programme for artificial-intelligence companies to submit cutting-edge models for US review.

The administration’s safety initiative accelerated after Anthropic warned in April that its Mythos model was adept at finding computer vulnerabilities and strictly limited its release. Adding urgency to the US effort, OpenAI and Anthropic have in recent weeks disclosed that some of their models escaped secure testing environments and hacked third-party organisations.

Debate in Washington over how to handle open-weight AI models has intensified following a wave of open systems from Chinese developers that rival top offerings from the US. The release in July of the new Kimi K3 model by Chinese start-up Moonshot also undermined investor confidence in the durability of the US’ lead in AI and stirred questions about the billions of dollars in data centre investments.

While some AI developers in the US, like OpenAI, offer open-weight models, much of their focus has been on closed or proprietary systems that they sell to customers. Chinese companies, including DeepSeek, have previously flooded the market with more affordable, open models that typically lag behind the most cutting-edge software from US rivals, but nonetheless offer a compelling product for the price.

Speaking at a cybersecurity conference in Las Vegas, National Cyber Director Sean Cairncross said that the administration is interested in supporting US open-source AI, saying that type of model has a “tremendous role. He said that the AI safety framework needs to have a “flexible” structure that encourages information sharing between the industry and government.  

“A regulatory regime would not only strangle growth, development and innovation, and be enormously harmful to the industry, but it would be obsolete 48 hours after it was going through whatever process it had gone through,” he said. 

At the same time, some officials in Washington have considered imposing restrictions on open-weight models, and the prospect of government curbs has drawn objections from across the tech industry. Silicon Valley leaders, including Nvidia chief executive officer Jensen Huang, have insisted that open-weight models are good for the long-term development of AI and help bolster its security.

The administration’s decision emerged less than two months before Trump is scheduled to meet his Chinese counterpart Xi Jinping for a high-profile summit in Washington, where the competition for AI leadership between the world’s two largest economies will take centre stage.

Exempting open-weight models from the opt-in safety testing programme marks a setback for Anthropic chief executive officer Dario Amodei, who has called for mandatory government safety reviews of both open and proprietary models. Amodei has also suggested that Chinese-made AI models violate US rules, concerns echoed by Treasury Secretary Scott Bessent and other top US officials.

In July, White House Science and Technology Policy Director Michael Kratsios accused Moonshot of illegally acquiring Nvidia’s advanced Blackwell chips, the most powerful AI accelerators on the market. It’s unclear how many of these processors the US government believes Moonshot has. Bloomberg News has reported that Moonshot has a computing agreement with Alibaba Group for the use of about 20,000 Nvidia chips, forming a substantial chunk of the overall capacity Moonshot uses for its Kimi models.

Kratsios also said Moonshot had improperly extracted data from US models through a process known as distillation, a style of reverse engineering that US AI companies have warned officials about for most of 2026. Using this technique, researchers feed the outputs of one model into a new version, a process that turns out to be an efficient way to recreate some of the parent model’s capabilities.

Partly in response to industry complaints of an ad-hoc US approach to model reviews, Bessent has proposed creating an independent regulatory agency that would give companies a significant say over safety reviews.

Bessent’s model would set up a system similar to the Financial Industry Regulatory Authority, but apply to the artificial intelligence industry, Bloomberg News has reported. It remains unclear if that plan will be part of the AI cybersecurity order implementation or a separate effort. BLOOMBERG