Governance & Compliance

Run the right model, on your hardware, in your jurisdiction.

Sovereign AI means controlling the system end to end, not taking a vendor’s word for which model is running or where it came from. VAIL verifies that the model on your hardware is the one you approved, traces its lineage, and keeps a timestamped record for audit.

Request an assessment What sovereign control requires

What sovereign control requires

Public sector teams are moving off third-party model APIs to open-weight models on infrastructure they control, often fine-tuned on data that cannot leave their network. That control holds only if you can show four things.

YOUR NETWORK · YOUR JURISDICTION Open-weight model hub · vendor LINEAGE Candidate model source selection IDENTITY Installed model your hardware BASELINE Fine-tuned model private data CHANGE Production serving requests VAIL Model Assurance runs inside your network · air-gap ready
Every check runs inside your network. Lineage is established before adoption, identity at installation, a baseline when you fine-tune, and change in production.

  • Your model, not a substitute

    The model on your hardware is the one you assessed and approved, and has not been swapped or altered since it was installed.

  • Known lineage

    Where an open-weight model came from, including what it was fine-tuned or distilled from. Lineage crosses borders: the Cisco study found NVIDIA Nemotron models post-trained from Alibaba Qwen base weights.

  • The right jurisdiction

    Evidence of where a model ran. In a joint study with Lucid Computing and Lunal, VAIL’s fingerprint was combined with trusted hardware and location attestation to verify a model’s location and identity.

  • Your fine-tune, on your network

    A model fine-tuned on private data is your most sensitive asset. VAIL fingerprints it at approval and monitors it for change, and can run air-gapped, so verification sends nothing outside your network.

How VAIL verifies it

01Source selection
Before you adopt a model, weight analysis and behavioral fingerprints establish its identity and lineage against a reference catalog of known models.
02Installation audit
When the model is installed, its behavioral fingerprint is matched to the one from source selection.
03Continuous monitoring
The running model is fingerprinted on a schedule, and any drift or change is flagged. Every stage runs on your hardware.
04Audit record
A timestamped record of model identity, lineage, and change events that audit and risk teams can use for frameworks such as the EU AI Act, NIST AI RMF, and SR 11-7.

Reference catalog updates are small vector files that can be moved across an air gap. How Model Assurance works

Built for

Teams
Public sector and sovereign AI programs, compliance officers, AI governance leads, internal audit, legal, and the auditors and insurers that assess AI systems.
Deployment
No SDK, no code changes, and no access to customer traffic. Managed, in your cloud, or air-gapped.

Research

All publications

Go further

Model Assurance Know which model is running, and where it came from.

Request an assessment.

Tell us which model or agent you need verified. We reply within one business day with a scope and a start date.

Or write to contact@projectvail.com

Reply within one business day