The Trump government is struggling to respond to the controversy surrounding a new Chinese model of artificial intelligence (AI), accused of plundering major American players but also hailed by supporters of cheaper and more open AI.
• Also read: Anthropic releases new AI, presented as less dangerous for cybersecurity
• Also read: Data center electricity consumption could quadruple by 2035
• Also read: How much of the music on Deezer is of AI origin? More than half!
On Wednesday, White House science and technology advisor Michael Kratsios said that the young, fast-growing Chinese company Moonshot AI had copied Anthropic’s Fable model, considered the most efficient on the market, to develop its Kimi K3 AI.
The latter has caused a sensation since its release a week ago, positioning itself ahead of the best American AI in several rankings, notably for the programming of sites and applications, a first for a so-called open (modifiable) Chinese model.
According to the advisor, Moonshot AI used the so-called distillation method, which consists of studying a model in detail from the outside, mainly by bombarding it with requests, to understand its functioning and trends.
Michael Kratsios echoed US Finance Minister Scott Bessent, who said the government has found traces of US models in several cutting-edge Chinese AI.
“If we conclude that foreign models are plundering our companies,” he said, “we will have the opportunity to sanction them. »
According to the site The Informationa specialized department of the Ministry of Commerce (BIS) is also investigating the use, by Moonshot AI, of chips from the American Nvidia theoretically prohibited for export to China.
A movement within the Trump government is campaigning, according to the news site Axiosfor several Chinese companies behind cutting-edge models to be placed on a BIS list restricting their commercial relations with American companies.
“Serious risks”
But some members of the American administration are opposed to this hard line, notably the Minister of Commerce, Howard Lutnick, again according to Axios.
In a message posted on
Like its internal dissensions, the American government is caught between two pressure campaigns.
On the one hand, a few American giants, worried about seeing their AI pillaged but also their economic model, based on the control of protected technology, called into question.
Illicit distillation “is intellectual property theft and industrial espionage,” Sarah Heck, head of external relations at Anthropic, told X. She sees it as “a national challenge that poses serious national security risks for the United States and allied democracies.”
But at the other end of the spectrum, many more are demanding a status quo to preserve the ecosystem of open AI, Chinese or not, accessible for free and adaptable according to needs.
On Wednesday, a group of 179 high-growth startups called the Little Tech Association sent a letter to the Trump administration advocating a measured approach. It would notably imply that constraints on the use of foreign models be applied only to government services.
On Friday, industry heavyweights including Microsoft, Mistral, Perplexity and Meta also called for restraint, even in cases of “illegal” distillation.
They propose addressing the issue through “a legal and commercial framework rather than broad restrictions”.
“These Chinese models are excellent,” Jensen Huang, boss of AI chip juggernaut Nvidia, told Axios, which is nevertheless seeking to popularize its own open model, Nemotron.
“Open models that are excellent should be able to be used. »
After several days of gestation, the debate seems to be turning in favor of open models, Chinese included, as shown by the reaction of Sam Altman, boss of OpenAI, to Friday’s call, of which his company was not a signatory. “I want the United States to win on both the open and closed models,” he wrote on X, “and I’m happy to see that.” »
