Anthropic accuses Alibaba of mass AI capability 'theft'

Artificial intelligence firm Anthropic has accused Chinese tech giant Alibaba of conducting a large-scale operation to illicitly extract capabilities from its Claude AI model. Anthropic alleges that Alibaba used "distillation attacks," a method where outputs from a sophisticated AI are used to train a less advanced model, to replicate Claude's advanced reasoning abilities.

In a letter to U.S. senators, Anthropic detailed that nearly 29 million interactions with Claude were carried out using thousands of fraudulent accounts allegedly linked to Alibaba. This campaign, described as the "largest campaign to extract Claude's capabilities illicitly," specifically targeted Claude's most advanced features, including its complex task handling and decision-making processes.

Anthropic, a key player in the global AI race, is urging the U.S. Congress to impose penalties on companies engaged in such practices. The company argues that these "distillation attacks" effectively turn billions of dollars in American investment into a subsidy for geopolitical competitors. Anthropic's claims echo previous accusations by OpenAI against Chinese entities and highlight growing concerns over the theft and replication of U.S. AI technology.

“

Distillation attacks turn hundreds of billions of dollars in American investment and [research and development] into a massive subsidy for our geopolitical competitors.