Anthropic Calls for Congressional Action Against AI Model Distillation by Chinese Entities

Anthropic Calls for Legislative Action

Anthropic is pressing Congress to strengthen measures against AI model distillation, alleging that entities linked to Alibaba executed a significant extraction campaign targeting its Claude chatbot. This appeal highlights growing concerns over unauthorized access to advanced U.S. AI technologies.

In a letter sent on June 10 to Tim Scott and Elizabeth Warren, key figures in the Senate Banking Committee, Anthropic detailed that Alibaba-affiliated operators conducted over 28.8 million exchanges with Claude between April 22 and June 5, using approximately 25,000 fraudulent accounts. This tactic, known as a distillation attack, aimed to siphon off Claude's advanced reasoning and planning capabilities, allowing competitors to mimic its performance without the substantial investment required for developing state-of-the-art AI systems.

“This campaign was notable not only for its scale but also for its audacity,” Anthropic stated, emphasizing the implications of such actions by a publicly traded company operating in the U.S.

Anthropic argues that this issue extends beyond intellectual property, framing it as a national security concern that could bolster China's military and cyber AI capabilities, thereby threatening U.S. technological supremacy. This call for action aligns with recent efforts by the U.S. government to protect its leadership in AI, following an executive order signed by President Donald Trump to enhance AI-related cybersecurity.

Anthropic contends that unauthorized extraction of capabilities from U.S. models allows foreign entities to benefit from American R&D investments without incurring associated costs, undermining the economic rationale for U.S. AI leadership. The company has called for enhanced intelligence sharing among AI developers, clearer antitrust regulations for information sharing regarding distillation attacks, tighter export regulations on high-level AI resources, and penalties for companies engaged in extensive model extraction.

While Anthropic's claims are met with skepticism by some industry observers, who note that similar techniques are often used in legitimate AI training, the company maintains that their practices violate terms of service and constitute unauthorized access. The discourse surrounding distillation has become increasingly intricate, particularly following testimonies from industry figures like Elon Musk, who acknowledged the use of models from other companies during training processes, blurring the lines between acceptable practices and unauthorized extraction.