Anthropic Distillation Claims Intensify AI Model Governance Risks

San Francisco, 11 September 2026 – Allegations that Chinese artificial intelligence developers used concealed routing and model-distillation techniques to extract capabilities from leading Western systems have widened an already consequential debate over intellectual property, access controls and national security. Anthropic has said campaigns linked to DeepSeek, Moonshot AI and MiniMax generated large volumes of interactions with Claude through intermediary accounts and infrastructure, with the apparent goal of improving rival models. The companies named have not been shown in the available material to have accepted those allegations, so investors should treat the claims as contested rather than settled findings.

Model distillation itself is a recognised engineering method. A smaller model can be trained to reproduce useful patterns from a larger system, lowering inference costs and accelerating deployment. The commercial dispute begins when developers allegedly evade contractual restrictions, disguise their identity or automate queries at a scale designed to reconstruct protected behaviour. In that setting, the issue moves beyond ordinary competition and into questions about misappropriation, cybersecurity and the enforceability of application programming interface terms across borders.