Editorial illustration for Cisco's AI Tool Fingerprints Nearly 900 Open Models, Expanding Free Verification Kit
Cisco AI Tool Fingerprints 900 Open Models Free
Every model repository on Hugging Face carries a base_model tag, a one-line claim about lineage typed in by whoever uploaded the file. Nobody checks it. The ATOM Report, published by Nathan Lambert and Florian Brand at Interconnects AI in April 2026, tracked roughly 1,500 mainline open models and found that Alibaba's Qwen family is now the declared parent of 69% of new open-model derivatives as of February 2026, up from 1% in January 2024.
Chinese labs collectively account for 70% of that lineage; Europe accounts for 4%. Across the tracked regions, cumulative downloads hit 2.04 billion by March 2026.
The problem sits underneath those numbers. A security team clearing a model for production reads a repository page listing a name, a license, and a base-model tag, none of which Hugging Face requires anyone to prove through actual weight analysis. Malware scanning has a similar hole: Hugging Face runs a ClamAV-based scanner and posts a badge per file, but the platform's own documentation says a file without an "ok" or "infected" badge could simply be unscanned, queued, or stuck in error. Coverage has been something reviewers assumed rather than something they could verify.
Cisco on Thursday published the AI Supply Chain Provenance Explorer, a free public database covering almost 900 open models. Each entry can carry provider headquarters, a fingerprinted lineage graph, license restrictions, and a files-scanned count.
Why this matters
For anyone pulling models off Hugging Face to ship a product, that lineage tag has always been a leap of faith. Cisco's expansion, covering nearly 900 models now versus 150 in April, makes weight-level verification something a security team can run without a GPU cluster or a week of engineering time. That matters because the ATOM Report's 69% figure isn't a rounding error.
It means most of what gets labeled "fine-tuned from Llama" or "based on Mistral" is an uploader's typed claim, not a checked fact. We'd treat this less as a nice-to-have compliance tool and more as a baseline check that should already be in every model-approval pipeline. The real test is adoption: will Hugging Face itself start surfacing this verification data on repository pages, or will it stay a separate tool people have to remember to run?
Cisco making it free removes the cost excuse. It doesn't remove the incentive problem, since uploaders benefit from vague lineage claims and nobody's forcing accountability yet. Watch whether platforms start requiring this kind of check before models get downloaded, not after.
Related Reading
Open Source D&B rebuilds 642 million‑business database after AI agents hit limits
Open Source OpenAI's GPT Transcribe Cuts Error Rate to 3.31%, Improving on GPT-4o
Open Source Study: 6.7% of Deepfake Requests on Hugging Face Targeted Children
From Archives Moonshot AI's MoonEP Uses Dynamic Redundant Experts to Balance MoE Training Load
From Archives OpenAI's AI Agent Ditched Its Exam to Chase Answer Key
Common Questions Answered
What is the AI Supply Chain Provenance Explorer and what does it cover?
Cisco's AI Supply Chain Provenance Explorer is a free public database that covers nearly 900 open models and provides fingerprinted lineage graphs, provider headquarters information, license restrictions, and file-scanned counts for each entry. This tool expands Cisco's previous offering which only covered 150 models in April, making it significantly more comprehensive for verifying model provenance.
Why is the ATOM Report's finding that 69% of open models trace back to Alibaba's Qwen family significant?
The 69% figure represents a dramatic shift in open model lineage, up from just 1% in January 2024, indicating that Alibaba's Qwen family has become the dominant parent model for new open-model derivatives. This concentration, combined with Chinese labs collectively accounting for 70% of that lineage, highlights a major shift in the open AI ecosystem that security teams need to understand when evaluating model provenance.
What problem does Cisco's fingerprinting solution solve regarding Hugging Face model tags?
Previously, the base_model tags on Hugging Face repositories were self-reported claims with no verification, making it a "leap of faith" for anyone using these models in production. Cisco's fingerprinting technology enables security teams to verify model lineage without requiring a GPU cluster or weeks of engineering time, replacing unverified uploader claims with cryptographic proof of model ancestry.
How has the declared parent model lineage changed from January 2024 to February 2026?
Alibaba's Qwen family went from being the declared parent of only 1% of new open-model derivatives in January 2024 to 69% by February 2026, representing a massive consolidation around this single model family. This shift demonstrates how quickly the open model landscape can change and why accurate lineage tracking has become critical for supply chain security.
Further Reading
- AI Supply Chain Provenance Explorer for Responsible AI - Cisco Blogs
- Cisco releases open-source 'DNA test for AI models' - SC Media
- Cisco Releases Open Source Tool for AI Model Provenance - SecurityWeek
- Cisco releases open-source toolkit for verifying AI model lineage - Help Net Security
- Cisco launches 'DNA Test' for AI models - The Deep View
