Washington's move to restrict foreign access to Anthropic's Claude models backfired within days. The export control, intended to protect American AI dominance, has accelerated competitors instead of slowing them. DeepSeek closed a $7.4 billion funding round on the heels of the announcement. Chinese AI labs are slashing API token prices up to 99 percent to capture fleeing customers. Cohere reports surging government interest from the exact regions Anthropic lost access to.
The strategy assumes cutting off Claude preserves American advantage. The opposite is happening. Customers blocked from Anthropic's models are switching to Chinese alternatives at fire-sale prices. DeepSeek's funding demonstrates investor confidence in China's ability to build competitive systems despite U.S. restrictions. Lower prices accelerate adoption, building larger user bases and generating more data to train the next generation of models.
Export controls work when they target scarce physical resources. AI models are software. They copy instantly. Restricting access to one provider simply redirects demand elsewhere, often to better-capitalized competitors. Anthropic loses revenue and market position. China's labs gain users and resources to improve their systems.
Separately, researchers found 144 poisoned npm packages designed to harvest credentials from AI supply chains. The packages mimicked legitimate machine learning libraries but contained malicious code. They targeted developers building AI systems, stealing API keys and authentication tokens. The attack exploited npm's size and developers' trust in open source ecosystems.
The incident reveals a second vulnerability in AI infrastructure. As companies race to deploy models, they pull from package repositories with minimal verification. A few lines of hidden code can compromise entire applications. The attack vector is trivial to execute but difficult to detect at scale.
These events outline two problems with current AI policy. First, restrictions on closed models drive adoption of open alternatives from rivals with fewer compliance incentives.