US artificial intelligence (AI) giant Anthropic has accused Chinese e-commerce and technology company Alibaba of ‘brazenly’ and ‘illegally’ extracting features from its Claude AI model.
In a letter to two U.S. members of Congress, the San Francisco-based company said Alibaba-linked operators conducted nearly 29 million exchanges with Claude using thousands of fraudulent accounts in what it called the largest extraction campaign of its kind.
Anthropic called on Congress to punish the companies behind attacks like this and strengthen measures to prevent the theft of American technology.
The BBC has contacted Alibaba for comment and asked Anthropic for further details.
Anthropic’s letter, dated June 10 and sent to U.S. Senators Tim Scott and Elizabeth Warren, accused New York Stock Exchange-listed Alibaba of carrying out “the largest campaign to illegally extract Claude’s capabilities.”
According to Anthropic, the campaign was conducted through a “distillation attack” that extracts answers from a stronger AI model to train a weaker AI model.
Alibaba-linked operators targeted Claude’s most valuable skills, including his ability to handle longer, more complex tasks and his approach to decision-making, Anthropic said.
These types of attacks are carried out at “industrial scale,” allowing Chinese companies to harvest and repackage U.S. AI capabilities for themselves, the company said.
The letter also mentions other alleged attacks that pose a threat to U.S. forces.
“Distillation attacks divert hundreds of billions of dollars in U.S. investment and research and development into massive subsidies for geopolitical competitors,” Anthropic said.
The report cited allegations from the U.S. Department of Defense that several major companies, including Alibaba, automaker BYD and technology company Baidu, have ties to the Chinese military.
The companies have denied the claims, and Alibaba sued the U.S. government this week to remove its name from the Pentagon’s blacklist.
US developers have previously accused Chinese competitors of using distillation attacks to train models that would compete with US AI technology at a fraction of the cost.
OpenAI has previously accused Chinese groups of adopting the same practices.
Anthropic is preparing for a blockbuster stock market debut with OpenAI, a leading AI developer and maker of ChatGPT, that could make it one of the world’s most valuable companies.
However, some of Anthropic’s more advanced models, such as Mythos, have raised cybersecurity concerns about their ability to target weaknesses in computer systems.
