Why 'Country of Origin' Labels Don't Tell the Whole AI Story

Many businesses trying to avoid AI tools linked to China are relying on simple country-of-origin labels to make that call. But new research from Cisco suggests this approach can be misleading. AI models are often built on top of other models through a process of adaptation and fine-tuning, meaning a tool marketed as being developed elsewhere may still carry the underlying code, behaviours, and potential vulnerabilities of an earlier Chinese-origin model.

This matters because these inherited traits aren't always obvious. A model's outward branding or the location of its developer doesn't guarantee where its core technology actually originated, or what security assumptions and weaknesses came bundled in along the way. For businesses making decisions based on national security or compliance concerns, this creates a blind spot: you could be excluding tools based on labels while unknowingly using others with the same underlying lineage.

Cisco's findings highlight a broader lesson for any organisation evaluating AI tools: transparency claims need to be checked against technical reality, not just marketing or country labels. As AI adoption grows among small and medium businesses, understanding where the technology actually comes from is becoming as important as understanding what it does.