tech
Did Alibaba Illegally Extract Anthropic’s AI Capabilities?
Anthropic claims Alibaba used thousands of fake accounts to distil Claude at scale, adding fresh fuel to the intensifying US-China race for AI dominance
TL;DR
- Anthropic alleges Alibaba conducted a large-scale distillation attack to steal capabilities from its Claude AI models.
- The attack reportedly involved over 28.8 million exchanges using approximately 25,000 fraudulent accounts between April and June 2026.
- Knowledge distillation is a legitimate ML technique that threat actors exploit to create cheaper replicas of proprietary AI models.
- This incident adds to the growing US-China competition for AI dominance, with Chinese models showing rapid advancements.
- Chinese AI startup Z.ai's model GLM-5.2 is noted for closely trailing Western frontier AI performance.
- Anthropic has previously accused other Chinese AI companies like DeepSeek, Moonshot AI, and MiniMax of similar extraction campaigns.
- The Pentagon has designated Alibaba as a Chinese military company, a status the company contests.