Cisco Fingerprinting Exposes Open Model Verification Gap

A recent report by Interconnects AI, known as the ATOM Report, has shed light on the lineage of open-source models. The report tracked approximately 1,500 mainline open models and found that 69% of new open-model derivatives claim to be based on Alibaba's Qwen family as of February 2026. This is a significant increase from January 2024, when the figure was only 1%. The report also notes that Chinese labs account for 70% of these models, while European labs account for only 4%. The verification of a model's lineage is crucial for security purposes, as it can help identify potential vulnerabilities. However, the current system relies on a tag populated by the uploader, which may not be accurate. The Cisco Foundation AI has taken steps to address this issue by scanning every public file uploaded to Hugging Face using an updated ClamAV engine. This effort has led to the fingerprinting of almost 900 models, providing a more transparent understanding of the open model landscape.

Why it matters

The lack of verified lineage in open models poses a significant security risk, as it can make it difficult to identify and address potential vulnerabilities. This issue highlights the need for more robust verification processes in AI development.