OpenAI blames Moonshot for mass data extraction on its AI models

Skip to Content News Archives Economy Energy Oil & Gas Renewables Electric Vehicles Mining Commodities Agriculture Real Estate Mortgages Mortgage Rates Finance Banking Insurance Fintech Cryptocurrency Work Wealth Smart Money Wealth Management Investor Personal Finance Family Finance Retirement Taxes High Net Worth FP Comment Executive Women Puzzmo Newsletters Financial Times Business Essentials More Innovation Information Technology FP500 Podcasts Small Business Lives Told Tails Told Shopping Financial Post Store Obituaries Place a Notice Advertising Advertising With Us Advertising Solutions Postmedia Ad Manager Sponsorship Requests Classifieds Place a Classifieds ad Working Profile Settings My Subscriptions My Offers Newsletters Customer Service FAQ News Economy Energy Mining Real Estate Finance Work Wealth Investor FP Comment Executive Women Puzzmo Newsletters Financial Times Business Essentials This advertisement has not loaded yet, but your article continues below.HomeInformation TechnologyNewsOpenAI blames Moonshot for mass data extraction on its AI modelsThe chatbot maker’s accusations highlight the intensifying global competition for OpenAI and other U.S. tech companiesAuthor of the article:Last updated 7 minutes ago Moonshot, whose breakthrough Kimi K3 model upended assumptions about Chinese AI capabilities, has faced particular scrutiny its rapid gains on U.S. rivals. Photo by Lam Yik/Bloomberg via Getty ImagesOpenAI accused its Chinese rival Moonshot AI of being responsible for a wide-scale effort to extract data from its GPT artificial intelligence systems that could be used to reproduce the reasoning and capabilities of the company’s most advanced models.THIS CONTENT IS RESERVED FOR SUBSCRIBERS ONLYSubscribe now to read the latest news in your city and across Canada.Exclusive articles from Barbara Shecter, Joe O'Connor, Gabriel Friedman, and others.Daily content from Financial Times, the world's leading global business publication.Unlimited online access to read articles from Financial Post, National Post and 15 news sites across Canada with one account.National Post ePaper, an electronic replica of the print edition to view on any device, share and comment on.Daily puzzles, including the New York Times Crossword.SUBSCRIBE TO UNLOCK MORE ARTICLESSubscribe now to read the latest news in your city and across Canada.Exclusive articles from Barbara Shecter, Joe O'Connor, Gabriel Friedman and others.Daily content from Financial Times, the world's leading global business publication.Unlimited online access to read articles from Financial Post, National Post and 15 news sites across Canada with one account.National Post ePaper, an electronic replica of the print edition to view on any device, share and comment on.Daily puzzles, including the New York Times Crossword.REGISTER / SIGN IN TO UNLOCK MORE ARTICLESCreate an account or sign in to continue with your reading experience.Access articles from across Canada with one account.Share your thoughts and join the conversation in the comments.Enjoy additional articles per month.Get email updates from your favourite authors.THIS ARTICLE IS FREE TO READ REGISTER TO UNLOCK.Create an account or sign in to continue with your reading experience.Access articles from across Canada with one accountShare your thoughts and join the conversation in the commentsEnjoy additional articles per monthGet email updates from your favourite authorsSign In or Create an AccountIn a blog post Wednesday, OpenAI disclosed that it observed thousands of attempts by users associated with Moonshot to decipher hidden information about how its models reason through problems. Coordinated efforts to glean the data began in early July, peaking at 16,000 requests later in the month. Though it couldn’t attribute all the operators to a single actor, OpenAI said that users associated with Moonshot played a significant role.This advertisement has not loaded yet, but your article continues below.OpenAI’s accusations against Moonshot add to a growing pile of claims from Silicon Valley and the Trump administration that Chinese AI developers are systematically extracting proprietary knowledge from American firms using a technique known as distillation to build a rival generation of chatbots at a fraction of the cost. Moonshot, whose breakthrough Kimi K3 model upended assumptions about Chinese AI capabilities, has faced particular scrutiny in Washington for its rapid gains on United States rivals.Breaking business news, incisive views, must-reads and market signals. Weekdays by 9 a.m.By signing up you consent to receive the above newsletter from Postmedia Network Inc.A welcome email is on its way. If you don't see it, please check your junk folder.The next issue of Posthaste will soon be in your inbox.We encountered an issue signing you up. Please try againWhile distillation is a common AI industry technique, where outputs from a state-of-the-art model are used to train a less-sophisticated one, the practice can violate developers’ terms of service when employed at an industrial scale. OpenAI, Anthropic PBC and Alphabet Inc.’s Google have been coordinating on ways to counter what they call adversarial forms by distillation — including by firms from China.“Our concern is about violation of our terms of service, not open models or legitimate distillation,” said Caroline Zier, lead for strategic national security policy initiatives at OpenAI. “This is also a shared challenge of ensuring the U.S. remains in the lead on all fronts, which is why we share findings with other developers and relevant government partners, and will also continue to invest in stronger protections. We support a thriving open-weight model ecosystem and the U.S. leading it.”This advertisement has not loaded yet, but your article continues below.Representatives for Moonshot didn’t immediately respond to requests for comment. China’s government has rejected previous distillation accusations from Silicon Valley firms and the Trump administration, warning that it would retaliate against any penalties imposed by the U.S. government.In part to limit damage from distillation attempts, OpenAI has been increasingly hiding its models’ reasoning chains, providing users with answers, rather than explanations of how it reached those answers. Individuals tied to Moonshot got around those restrictions, OpenAI’s blog says, by copying encrypted reasoning from one conversation and asking a model in another conversation to transcribe it.The method involved a narrow jailbreak to reveal a version of the reasoning legible to the company’s internal servers and did not decrypt it, OpenAI said. The evolving nature of the attacks as OpenAI changed its defences is evidence that such distillation provides valuable training, according to a person at an AI lab, who asked not to be named for confidentiality reasons.This advertisement has not loaded yet.This advertisement has not loaded yet, but your article continues below.OpenAI’s accusations highlight the intensifying competition for OpenAI and other American companies, which are grappling with a growing challenge from Chinese rivals including DeepSeek and Minimax that offer more affordable open-weight models. While some AI developers in the U.S., including OpenAI, offer open systems, much of their focus has been on closed or proprietary systems that they sell to customers.A few weeks after Kimi K3 launched, White House science and tech policy advisor Michael Kratsios said in a social media post that Moonshot had accessed Nvidia Blackwell computing servers, which are banned for sale to Chinese companies, and used a “sophisticated internal platform” to extract data from U.S. models.Moonshot has a computing power agreement with Alibaba Group Holding Ltd. for the use of around 20,000 Nvidia Corp. chips, a cluster that makes up a substantial chunk of the overall computing capacity it uses for Kimi models, Bloomberg News has reported.The disclosures mark the first time OpenAI has accused Moonshot, but it follows a threats report from Anthropic last month that said Moonshot covertly routed thousands of user requests to the U.S. firm’s Claude models and passed off the responses as its own, while using the answers to help train Kimi models.This advertisement has not loaded yet, but your article continues below.OpenAI’s latest disclosure of a distillation attempt involving 16,000 requests is only a fraction of the size of the largest incidents reported by Anthropic, which recorded distillation in the millions of exchanges.A person familiar with OpenAI’s thinking, who also spoke on condition of anonymity for confidentiality reasons, cautioned that the figures may cover different time periods and use different detection methods, so they may not be directly comparable. If an attack is disrupted early, there will also be less volume overall, the person added.Distillation accusations by U.S. companies have their critics, including David Sacks, a venture capitalist and President Donald Trump’s former AI czar. Sacks has characterized the reports as an attempt to get the US to ban open models that challenge OpenAI and Anthropic’s business model. China has also fervently opposed claims that its companies are engaging in unfair development practices.This advertisement has not loaded yet, but your article continues below.OpenAI has also sought clarity on antitrust law to allow developers to share safety and security related information and wants open-weight developers to participate, the people added. Senator Jim Banks, an Indiana Republican and Senator Adam Schiff, a California Democrat, have proposed legislation that would create a narrow antitrust exemption for sharing information related to unauthorized distillation and other AI safety threats.To some extent, OpenAI has already been sharing information with others that might face similar distillation attempts, and it said in the blog that it had told industry partners about this incident through the Frontier Model Forum. It also communicated it to government channels, a pathway blessed by a national security policy memo signed by Kratsios in April.—With assistance from Rachel Metz.We apologize, but this video has failed to load.This advertisement has not loaded yet.Notice for the Postmedia NetworkThis website uses cookies to personalize your content (including ads), and allows us to analyze our traffic. Read more about cookies here. By continuing to use our site, you agree to our Terms of Use and Privacy Policy.