Anthropic Tells Senate Alibaba Ran the Largest Known AI Distillation Attack on Claude — 25,000 Fake Accounts, 28.8M Exchanges
One-line summary: Anthropic formally accused Alibaba's Qwen lab of orchestrating a six-week campaign using ~25,000 fraudulent accounts and proxy services to systematically copy Claude's coding and agentic-reasoning capabilities.
Key facts
- Timeline: April 22 – June 5, 2026 (~6 weeks)
- Scale: ~25,000 fraudulent accounts; 28.8 million Claude exchanges — the largest distillation attack Anthropic has ever publicly documented
- Method: commercial proxy services used to bypass geographic access restrictions barring Chinese entities from Claude
- Target: Claude's software-engineering and agentic-reasoning capabilities
- Anthropic's June 10 letter addressed Senate Banking Committee Chair Tim Scott (R) and Ranking Member Elizabeth Warren (D)
- Dwarfs the combined scale of three Chinese AI labs Anthropic publicly named in February 2026
Why it matters
This is the most concrete example yet of a frontier AI lab formally accusing a named rival of industrial-scale model distillation. If confirmed, it shows that API access controls can be circumvented at scale through coordinated proxy operations — and will likely accelerate regulatory and technical pressure on authentication, rate limiting, and behavioral monitoring at model providers across the industry.