Anthropic Urges Congress to Crack Down on AI Distillation By Chinese Rivals
Anthropic urged Congress to tighten protections against AI model distillation, alleging that operators linked to Alibaba and its Qwen lab used nearly 25,000 fraudulent accounts to generate over 28.8 million Claude exchanges from April 22 to June 5. It said the activity targeted Claude’s reasoning, coding, and long-horizon planning abilities, enabling rivals to copy frontier-model behavior without paying training costs. Anthropic framed the issue as both intellectual property theft and a national security risk, arguing it could speed up China’s military and cyber AI development and erode the U.S. lead. It asked lawmakers to expand intelligence sharing on distillation attacks, clarify antitrust rules for information sharing, tighten export controls on AI chips and compute, close access loopholes to overseas data centers, and impose penalties for large-scale model extraction. The letter follows earlier claims of similar activity by other Chinese labs and comes amid wider debate over where legitimate model training ends and unauthorized extraction begins.
