Overview
- Anthropic has publicly accused competitors of using 'distillation'—training new models on outputs from its Claude system—to harvest advanced capabilities and shortcut billions in development costs.
- Those claims have led major labs to tighten or block access to top models and have prompted U.S. regulators to impose selective gating and negotiate technical safeguards for vetted users.
- Microsoft CEO Satya Nadella pushed back on public complaints about distillation, calling them hypocritical and urging companies to build their own AI infrastructure and learning loops.
- Legal experts and journalists note symmetry between distillation and past web scraping by AI firms, leaving the legality of reusing model outputs uncertain and setting up competing fair‑use claims.
- The access curbs and legal uncertainty are already reshaping the market by driving some enterprises toward low‑cost Chinese open‑weight models and forcing firms to reassess data governance and vendor dependence.