Anthropic has accused Alibaba of running a campaign to “illicitly” extract the capabilities of its Claude AI models.
The Claude developer says operators affiliated with Alibaba and its Qwen AI lab generated more than 28.8 million exchanges with Claude through almost 25,000 fraudulent accounts.
It made the claim in a letter dated June 10 to two US senators, sent ahead of a Senate Banking Committee hearing the following day.
Anthropic’s letter arrives amid ongoing legal action the AI giant is facing as a defendant in the music industry’s copyright fight.
Universal Music Publishing Group, Concord, and ABKCO are suing it in two cases, including a January 2026 claim covering more than 20,000 songs and seeking over $3 billion, and BMG has filed a third.
All allege Anthropic copied song lyrics without a license to train Claude, which the company is contesting as fair use.
Anthropic’s allegation against Alibaba centers on a company with a foothold in music technology: Alibaba’s Qwen models can analyze recorded music.
“Alibaba executed the largest known distillation attack on Anthropic to date,” the company wrote in the letter.
The campaign ran between April 22 and June 5 and targeted what it called Claude’s most valuable capabilities, including agentic reasoning, software engineering, and long-horizon tasks, Anthropic said.
“These distillation attacks are carried out illicitly, systematically, and at industrial scale to harvest US AI capabilities across frontier labs and repackage them as their own,” Anthropic wrote.
“These distillation attacks are carried out illicitly, systematically, and at industrial scale to harvest US AI capabilities across frontier labs and repackage them as their own.”
Anthropic
The labs do so “without incurring the training and R&D costs required to train US frontier models,” Anthropic added.
“Distillation attacks turn hundreds of billions of dollars in American investment and R&D into a massive subsidy for our geopolitical competitors,” Anthropic said.
Anthropic’s spokesperson would not go into detail, telling CNBC only that countering distillation needs “coordinated action between government and industry.”
Distillation is a technique in which a less capable AI model is trained on the outputs of a stronger one to copy its behavior.
It can slash the cost of building a model, but Anthropic and its rivals say using it to clone a frontier system without permission breaches their terms of service.
Anthropic also warned that models built this way are often released with weak safeguards.
In a blog post in February, Anthropic named DeepSeek, Moonshot, and MiniMax, saying the three Chinese labs together generated more than 16 million exchanges with Claude through 24,000 fraudulent accounts.
The Alibaba campaign generated nearly twice as many exchanges as those three labs combined, on Anthropic’s figures.
Anthropic’s rivals have made similar accusations.
OpenAI said in January 2025 that China’s DeepSeek may have used its data to train a rival model without permission.
DeepSeek said it spent about USD $5.6 million training its model, against the more than $100 million OpenAI spent on GPT-4.
Its debut wiped roughly $600 billion off chipmaker Nvidia’s market value in a single day.
Anthropic, OpenAI, and Alphabet’s Google have since agreed to share information about distillation attempts that breach their terms of service, according to Bloomberg.
US officials have estimated that unauthorized distillation costs Silicon Valley labs billions of dollars, according to Bloomberg.
In the letter, Anthropic asked Congress to “facilitate threat information sharing between US AI labs, close loopholes allowing PRC AI labs to access advanced US chips, and penalize PRC labs responsible for distillation attacks.”
The first request reflects Anthropic’s view that antitrust rules currently make it harder for US labs to compare notes on the tactics they face.
The campaign ran after the Trump administration had already flagged the practice, Anthropic noted.
In April, White House science chief Michael Kratsios published a memo on curbing the exploitation of US model outputs, distinguishing it from legitimate research by its “industrial scale.”
The administration’s own science office had called such campaigns “unacceptable,” Anthropic noted.
Anthropic said PRC labs evade detection using proxy networks supplied by a “growing circumvention economy.”
Sens. Bill Hagerty and Andy Kim are preparing a defense-bill amendment that would penalize Chinese companies caught siphoning US models’ output, Bloomberg reported.
The request lands while Anthropic is itself at odds with the same administration.
The Trump administration this month imposed export controls on Anthropic’s top two models, Fable 5 and Mythos 5, on security grounds.
Anthropic, now valued at $965 billion and preparing for an IPO, said it disabled access to those systems more than a week ago.
That valuation lifted Anthropic above rival OpenAI for the first time.
OpenAI, valued at $852 billion, filed for its own IPO a week after Anthropic did the same.
AI built in China is already embedded in music creation.
Tencent Music Entertainment has integrated DeepSeek into its music-creation tools, letting users generate tracks and publish them directly to streaming app QQ Music.
Alibaba’s Qwen line can caption a song’s genre and mood, and its newest Qwen3.5-Omni model generates speech and clones voices.
On Alibaba’s own benchmarks, that model outperformed ElevenLabs on multilingual voice stability.
The backdrop is a flood of synthetic music: streaming service Deezer has said it receives 75,000 fully AI-generated tracks a day.
Music Business Worldwide
