Everyday AI Podcast – An AI and ChatGPT Podcast Ep 720: China Stealing AI from the U.S.? Inside Anthropic's Bombshell Allegations
54 snips
Feb 24, 2026 A deep dive into allegations that Chinese labs used millions of prompt exchanges to copy a major AI model. Discussion covers named companies, how distillation attacks work, and why token migration can dent U.S. AI revenues. The conversation also explores why technical defenses and export controls may fall short and the broader economic and geopolitical fallout.
AI Snips
Chapters
Transcript
Episode notes
Anthropic's Distillation Allegation Has Broad Economic Stakes
- Anthropic alleges three Chinese labs generated over 16 million exchanges with Claude using ~24,000 fake accounts to distill its capabilities.
- Jordan Wilson frames this as a national economic and strategic threat beyond just tech rivalry, affecting US companies and retirement portfolios.
Distillation Can Shortcut Model Development
- Distillation trains a smaller model on outputs from a stronger one, which can be legitimate internally but illicit when used by competitors.
- Anthropic warns competitors can acquire capabilities far faster and cheaper by distilling outputs from frontier models.
Evidence Claims Include IP Traces And Large Proxy Networks
- Anthropic traced Minimax, Moonshot, and DeepSeek activity via IP and metadata and says Minimax rerouted nearly half its traffic to capture a new system's capability.
- One proxy network reportedly managed over 20,000 fake accounts simultaneously to evade detection.