Anthropic has asked Congress to crack down on AI distillation by Chinese rivals, alleging that Alibaba-affiliated operators used nearly 25,000 fraudulent accounts to generate roughly 28.8 million Claude exchanges. The company claims this technique, known as distillation, lets competitors cheaply replicate proprietary models at scale, undercutting US AI leadership.
Distillation involves using a powerful AI model to train a cheaper copy, often through unauthorized API calls or fake user accounts. Anthropic's accusation gives Congress new ammunition for export controls and regulatory barriers against Chinese AI firms, though the exact technical and financial impact remains unclear from public disclosures.
Current US regulation lacks specific protections against model theft via fraudulent API consumption, leaving firms with recourse only through terms-of-service lawsuits. The Federal Trade Commission has flagged deceptive AI practices, but no targeted ban exists on distillation as a trade secret violation.
Anthropic's market cap is private, but its valuation exceeded $18 billion in 2023. The firm competes with OpenAI and Google DeepMind, where closed-source models like GPT-4o and Gemini are similarly vulnerable. Distillation poses systemic risk to US AI investment, as copycat models could erode pricing power and returns on R&D.
Critics argue the claims lack independent verification, noting Anthropic has financial incentive to portray rivals as threats to secure government contracts. Some analysts suggest distillation is a natural part of open-source AI evolution, and banning it could stifle global competition and innovation without tangible security benefits.