Nvidia and Palantir urge US not to ban ‘open’ AI models after China scare

{"text":[[{"start":10,"text":"Leading tech groups including Nvidia, Palantir, Microsoft and Meta have urged the US not to restrict “open-weight” AI models, as Washington threatens to target Chinese labs over fears of intellectual property theft."}],[{"start":22.25,"text":"Open models, which users can customise and run on their own hardware, are popular among Chinese AI developers who have recently announced technical advances against US rivals whose top-tier models are closed. "}],[{"start":35.7,"text":"The open letter on Friday, which argued policymakers should keep “the [AI] frontier plural by avoiding premature restrictions on open models”, was not joined by OpenAI, Anthropic or Google — the three leading makers of closed AI models. "}],[{"start":52.650000000000006,"text":"The public plea comes amid an intensifying fight over the fate of open-weight models. Calls for a ban on Chinese AI models have grown louder this month after the release of Kimi K3 by Beijing-based Moonshot AI, whose capabilities stoked concerns about China edging ahead in an AI arms race both countries regard as critical to national security and prosperity."}],[{"start":76.55000000000001,"text":"Michael Kratsios, director of the White House Office of Science and Technology Policy, this week said the US “has information” that Moonshot used a process called distillation — which involves training a less powerful model using the outputs from a stronger one — to develop its technology from Anthropic’s latest Fable model. "}],[{"start":95.10000000000001,"text":"“Large-scale, covert industrial distillation aimed at stealing proprietary US technology and undermining American research is unacceptable,” he said. "}],[{"start":104.35000000000001,"text":"In response to Kratsios’s claim, Anthropic said “industrial espionage . . . creates serious national security risks for the United States and democratic allies”."}],[{"start":115.75000000000001,"text":"US Treasury secretary Scott Bessent fuelled speculation about a ban on Chinese models, saying that “when PRC firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions . . . will be on the table”."}],[{"start":131.5,"text":"But he added that the Trump administration supports open-weight AI. "}],[{"start":null,"text":"

"}],[{"start":136.8,"text":"The availability of high-performing, openly accessible Chinese models is also a threat to the business models of the Silicon Valley groups. OpenAI and Anthropic have both attained valuations of close to $1tn this year, with investors betting the pair can sustain a lead over foreign rivals which will enable them to charge a premium for their models. "}],[{"start":156.9,"text":"Friday’s letter argued the availability of open models would also save businesses from “paying frontier-model prices for every task” and spread AI adoption more widely."}],[{"start":167.3,"text":"Investors including Donald Trump’s former AI tsar David Sacks have accused Anthropic of trying to quash open-source rivals by amping up the threat they pose in order to cement their competitive advantage. "}],[{"start":178.95000000000002,"text":"A preliminary assessment of Kimi K3’s capabilities by US and UK researchers released on Thursday found it was still way behind leading American models. US open-weight models are further behind still."}],[{"start":192.55,"text":"In an interview with Axios this week, Nvidia chief executive Jensen Huang said open-source models, including Chinese ones, should be embraced by US companies. “These Chinese open-source models are excellent,” he said, adding: “Great models lead to more usage.”"}],[{"start":208.55,"text":"Huang has also pushed the US to allow Nvidia to sell its AI chips, which are used to train the most advanced models, in China, arguing it would strengthen America’s leadership of the industry."}],[{"start":220.15,"text":"In Friday’s letter, the companies calling for open-weight models — which include venture capital group Andreessen Horowitz, IBM and Reflection — conceded these “carry real and distinct risks” as, once released, users can modify the models, giving creators little control over their ultimate use. "}],[{"start":237.5,"text":"However, they argued blocking their use would only increase such risks, whereas releasing them would allow “a broad community of researchers and developers to examine their behaviour, identify vulnerabilities, develop safeguards and improve them over time”."}],[{"start":261.05,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1784935304_4600.mp3"}