Don't trust the PDF. Verify the model.

Procurement-ready Model Origin Assurance that flags PRC-derived LLMs (Qwen family) and delivers a one-page Attribution Report your lawyers can staple to the deal.

✓ 72-hour SLA ✓ On-prem CLI ✓ API/SDK ✓ AI-BOM ready

Sample Attribution Report

One-Page Attribution Report

Clear, procurement-ready documentation that your legal team can review in under 5 minutes. Includes confidence scoring, detection methods, and supporting exhibits.

  • Executive summary with risk assessment
  • Technical findings and confidence scores
  • Supporting evidence and methodology
  • Legal compliance recommendations

How It Works

Our multi-layered detection system uses 4 independent verification methods

🏷️

Artifact Analysis

Scans configuration files, metadata, and model signatures for telltale patterns from Qwen family models.

🏗️

Architecture & Tokenizer

Analyzes model structure, layer configurations, and tokenizer vocabularies for architectural fingerprints.

⚖️

Weight Similarity

Compares weight distributions and parameters against known PRC model baselines using advanced statistical methods.

🧪

Behavioral Probes

Tests model responses to diagnostic prompts that reveal training data patterns and behavioral signatures.

Detection Confidence Levels

≥60 points HIGH - Likely PRC-derived
40-59 points MEDIUM - Possible lineage
20-39 points LOW - Some indicators
<20 points VERY LOW - Unlikely

Transparent Pricing

No hidden fees. Pay per report or choose a package that fits your needs.

Single Report

€2,500

Perfect for one-time verification

  • 72-hour delivery SLA
  • Full Attribution Report
  • Supporting exhibits
  • Email support
Get Started

Annual License

€60,000

300 reports/year - €200 each

  • On-premises deployment
  • API access included
  • Custom integrations
  • White-label reports
Contact Sales

Enterprise & OEM: Starting at €150,000/year for high-volume processing and custom deployment. Contact us for tailored solutions.

Security & Compliance First

🔐 On-Premises Deployment

Your models never leave your infrastructure. Run our analysis tools locally with full control over your IP.

🗑️ 7-Day Auto-Deletion

Any temporary analysis files are automatically deleted after 7 days. No long-term data retention.

🛡️ Encrypted Processing

All analysis bundles use end-to-end encryption. Your data is protected throughout the verification process.

📋 Compliance Ready

Reports designed for SOC2, GDPR, and enterprise compliance requirements. DPA available on request.

Frequently Asked Questions

How accurate is your detection?

Our multi-method approach achieves high accuracy by combining artifact matching, architecture analysis, tokenizer fingerprinting, weight similarity, and behavioral probing. We continuously update our signatures as new models are released.

Do you upload our model weights?

No, we never upload your model weights by default. Our on-premises solution keeps your IP secure while performing comprehensive analysis. Only metadata and configuration files are analyzed remotely if needed.

What's considered PRC-derived?

Models derived from Chinese AI models, particularly the Qwen family (Qwen, Qwen2, Qwen2.5) and their fine-tuned variants or architectural derivatives. This includes models that use Qwen as a base for further training.

Can this be automated?

Yes! Our API and CLI tools integrate into your procurement pipeline. Set up automatic scans for new model acquisitions or periodic audits of your model inventory.

What inputs do you need?

We can analyze HuggingFace model repositories, local model files (.safetensors, .bin), configuration files, or just model metadata depending on your security requirements.

How fast is the analysis?

Standard turnaround is 72 hours. Urgent requests can be processed in 24 hours for an additional fee. API-based quick scans provide preliminary results in minutes.

Get Your Free Model Assessment

Start with a complimentary analysis to see how our verification works. No commitment required.