The geopolitics of AI got sharper this week. Anthropic publicly accused Alibaba of conducting large-scale, illicit extraction of its Claude AI's capabilities — and took the unusual step of notifying U.S. senators.
The allegation
According to Anthropic, the activity involved nearly 29 million fraudulent interactions spread across thousands of accounts — a pattern the company characterizes as a deliberate effort to extract model capabilities at scale rather than ordinary usage. "Model extraction" refers to systematically querying a model to replicate its behavior or distill its knowledge into a competing system, a practice that sits in a legal and ethical gray zone the industry is still mapping.
Why notify lawmakers
By escalating to U.S. senators, Anthropic frames the issue as more than a commercial dispute. It places model security alongside national competitiveness and export-control debates, in which the capabilities baked into frontier models are increasingly treated as strategic assets worth protecting.
Why it matters
As frontier models become more valuable and harder to build, the incentives to copy them grow accordingly. The episode highlights an emerging front in the AI race: defending models not just from misuse, but from extraction. Expect tighter rate limits, more sophisticated abuse detection, and a louder policy conversation about where competitive intelligence ends and illicit extraction begins.