US artificial intelligence leader Anthropic has formally accused Chinese technology giant Alibaba of conducting an extensive campaign to illicitly extract capabilities from its Claude AI model. The San Francisco-based firm described it as the largest known "distillation attack," involving nearly 29 million interactions through thousands of fraudulent accounts. This serious allegation intensifies the ongoing debate over intellectual property protection and technological competition in the rapidly evolving AI landscape.
A Coordinated Campaign
In a detailed letter sent to the US Senate Banking Committee on June 10, Anthropic outlined the scale of the operation. The company stated that between late April and early June, operators linked to Alibaba used approximately 25,000 fake accounts to engage with Claude. The campaign specifically targeted the model's most valuable functions, such as agentic reasoning, software engineering, and its ability to handle complex, long-horizon tasks.
The Threat of AI Distillation
The technique, known as a distillation attack, involves using the outputs from a powerful AI model to train a weaker or newer one, effectively transferring its knowledge. While distillation is used widely in the industry for legitimate purposes, Anthropic and US officials frame this as industrial-scale theft. They argue it provides a massive subsidy for geopolitical competitors, allowing them to replicate billions of dollars in American research and development.
A Pattern of Behavior
This is not the first time Anthropic has raised such alarms, having previously accused other Chinese AI labs like DeepSeek, Moonshot, and MiniMax of similar illicit activities. The company noted that Alibaba's campaign followed the same pattern but was significantly larger in scale. OpenAI, the creator of ChatGPT, has also reported that Chinese groups have engaged in the same behavior to train their own models.
Geopolitical and Security Implications
The accusation lands as Alibaba is already under significant pressure from the US government, which has placed it on a Pentagon blacklist for alleged ties to China's military. Alibaba has strongly denied these claims and recently filed a lawsuit to have its name removed from the list. This new charge from a prominent AI firm complicates Alibaba's position and adds fuel to Washington's national security concerns.
These security fears are particularly acute regarding Anthropic's advanced Mythos model, which has sophisticated cyber capabilities. The US government has been collaborating with Anthropic to apply this model in cyber-offensive operations against adversaries. The potential for such technology to be replicated by a strategic competitor through distillation attacks represents a significant national security risk for the United States.
A Call for Stronger Regulations
In its communication with Congress, Anthropic issued a direct call for legislative action to protect American innovation. The company urged lawmakers to "close loopholes allowing PRC AI labs to access advanced US chips" and to penalize Chinese labs found responsible for these attacks. A White House official echoed these concerns, stating that industrial distillation of US AI models was "unacceptable" and that the administration intends to hold bad actors accountable.
Anthropic's bold accusation against Alibaba marks a critical escalation in the tech cold war between the US and China. The incident highlights the vulnerabilities of open AI platforms and the urgent need for robust safeguards against intellectual property theft. While Alibaba has yet to comment on these specific allegations, the episode is poised to influence future US policy on AI security and international technology transfers.