Anthropic just caught Alibaba running 28.8 million extraction exchanges against Claude. US sanctions are next.

Alibaba's Qwen AI lab ran 28.8 million exchanges through 25,000 fake accounts against Anthropic's Claude between April and June.

Not to test the model. To copy it.

This is adversarial distillation at industrial scale.

A smaller model trains on a frontier model's outputs until it replicates the capabilities at a fraction of the cost.

Anthropic's letter to the Senate Banking Committee says Alibaba targeted Claude's most commercially valuable skills — software engineering and agentic reasoning.

The same capabilities enterprises are paying premium prices to access.

Senators Hagerty and Kim are moving to amend must-pass defense legislation to blacklist or sanction any Chinese firm caught extracting US AI model outputs.

The Pentagon already added Alibaba to its military companies blacklist on June 8.

Your AI vendor's best models are being systematically replicated by competitors who skip the R&D entirely.

Models built through adversarial distillation lack safety guardrails.

That means the cheap knockoffs circulating in Chinese markets are not just IP theft — they are uncontrolled intelligence.

Audit your AI vendor relationships today.

If your workflows depend on frontier model capabilities, ask your provider what they are doing to detect and prevent distillation attacks.

This is not a vendor problem. It is your problem.

SOURCE: https://www.cnbc.com/2026/06/24/anthropic-alibaba-distillation-campaign.html

VERIFIED: CNBC, Bloomberg, The Next Web, Cybersecurity News, Business Times

SIGNAL: US-China AI decoupling just escalated from export controls to active IP warfare. Enterprise AI strategies that assume cross-border model access are now a compliance risk.