Anthropic’s Alibaba claim tests its AI edge ahead of IPO
Anthropic is asking Washington to respond after alleging Alibaba used Claude access to train rival AI, raising investor questions before a possible IPO.
By Hana Yoshida · Markets Reporter
3 min read
Anthropic has accused Alibaba of using access to Claude to help build competing AI systems, a claim that could shape how investors judge Anthropic’s durability before a potential public offering. The dispute matters because the company’s value depends in part on whether its most advanced models can stay ahead of rivals in China.
According to Fortune, Anthropic alleges Alibaba used fake accounts and ordinary-looking exchanges with Claude to draw out model behavior and train rival systems more cheaply. The technique at issue is known as distillation, in which one model’s outputs are used to improve another.
Anthropic has asked Washington for help. Sarah Heck, the company’s head of policy, urged Congress to respond to Chinese conduct through export controls on advanced American computing, according to a letter cited by Fortune.
Export rules face a gap
Kevin Wolf, a former assistant secretary of commerce for export administration, told Fortune that existing U.S. export controls are built mainly around hardware, such as chips, and certain direct transfers of software. He said access through an application programming interface does not amount to exporting the model itself.
That distinction creates a policy problem for Anthropic. Fortune reported that current controls can restrict some foreign access to tangible AI tools, including models such as Mythos and Fable, but they do not directly cover the kind of large-scale unauthorized distillation Anthropic says occurred.
The Trump administration addressed the issue in an April memo, according to Fortune. The memo called alleged efforts by Chinese companies to distill U.S. frontier models unacceptable, and Wolf said Anthropic’s allegations could add momentum to proposals for changing export-control rules.
One proposal cited by Wolf is the Remote Access Security Act, introduced last year by Rep. Michael Lawler, a New York Republican. According to Congress records cited by Fortune, the bill remains in committee and would target foreign access to U.S. technology through cloud computing services when use of the item could create a serious national security risk.
Lawler told Fortune that Anthropic’s capabilities should not reach China or other bad actors. He said his bill is meant to close a loophole that lets adversaries reach sensitive technology through the cloud.
Fortune reported that the bill would also address a gap left after the Trump administration rescinded a Biden-era framework that would have limited China’s access to AI cloud compute and model weights shortly before it was set to take effect.
IPO questions
Jay Ritter, a University of Florida IPO expert, told Fortune the Alibaba dispute could cut two ways for Anthropic. He said it could strengthen Anthropic’s case as a strategically important U.S. AI company, or it could cause investors to worry that rivals can erode its lead and future profits.
Ritter said the profitability concern would likely weigh more heavily with investors. He told Fortune that Anthropic’s revenue has grown rapidly, but its ability to maintain that growth remains uncertain.
Harrison Rolfes, a senior research analyst for private companies at PitchBook, told Fortune that a copied model could also underscore Claude’s value as the original system. He said many enterprises, especially U.S. companies, may still prefer a trusted frontier model even if cheaper Chinese alternatives are available.
Fortune reported that Anthropic is expected to pursue an IPO later this year that could value the company at $1 trillion. Rolfes said the company must balance its push for government protection against the risk that too much regulation could constrain its business.
This story draws on original reporting from Fortune.