US lawmaker presses OpenAI, Anthropic and three rivals on model weight theft

The letters also ask each company to describe the cybersecurity measures it uses to stop that kind of theft, according to The Next Web. Khanna represents part of Silicon Valley and is the top Democrat on the House Select Committee on China.

"The theft of such a model weight by (China) could erode America's AI lead with the stroke of a keyboard," he wrote, according to Reuters. The companies and the Chinese embassy in Washington did not immediately respond to Reuters' requests for comment.

What model weights actually are

Model weights are the numbers learned during training that make an AI model work. Anyone who holds them can run a copy of the model. That is why they sit at the centre of the current argument between US labs and their Chinese rivals. The same day, Khanna wrote to three Chinese labs, DeepSeek, Alibaba and Moonshot AI, and sent a further letter to the US intelligence community asking how they handle the risks of advanced AI.

"You cannot trust Silicon Valley tech billionaires to write the rules to keep us safe," Khanna told Reuters.

The timing is not accidental. OpenAI said on 1 October that it had identified and disrupted a coordinated campaign to extract protected reasoning from its models. It linked a core cluster of that activity to Moonshot AI, the developer of Kimi, CNBC reported. The activity began in early July and later surged to 16,000 requests from more than 4,000 users over two days. OpenAI said it fully disrupted the campaign by 28 July.

OpenAI called the operation "adversarial distillation", a technique in which one model's outputs or reasoning are used to train or improve another. The company said the operators did not breach its encryption, databases or stored user conversations. Instead, they manipulated model interactions so that hidden reasoning could be reproduced in a form visible to the requester. OpenAI said it was unclear whether all operators were linked to a single actor. It had shared findings through the Frontier Model Forum and government information-sharing channels. Moonshot did not immediately respond to CNBC's requests for comment.

The sceptical reading

The Register was blunter about it on 30 September, noting that OpenAI "hoovered up vast amounts of internet content amid copyright fights" while now describing distillation of its own outputs as a national security risk. The same report cites Michael Kratsios, US President Donald Trump's Assistant for Science and Technology, who in late July accused Moonshot AI of building its Kimi K3 model by distilling Anthropic's Fable. Anthropic's Claude Opus 5.5, released a week ago, ships with a defence against distillation called "preserved thinking".

How much of this is proven is a separate question. OpenAI and Anthropic have both accused Chinese firms, including Moonshot AI and DeepSeek, of distillation. But few cases of stolen model weights are publicly known, The Next Web notes.

Attention is also moving to policy. At OpenAI's developer day on 30 September, Sam Altman said the company would not go public until it can "make confident safety decisions", with a listing potentially valuing the firm at up to $1.4tn, the BBC reported. Trump, meeting tech leaders the same day, repeated that AI companies should regulate themselves.