How Does InferTrust™ Handle Model Versioning?

Learn how InferTrust™ tracks and cryptographically binds AI model versions to every decision record for complete model accountability.

How Does InferTrust™ Handle Model Versioning?

Every InferTrust™ (Patent Pending) decision record includes a Model Version ID (MID), a cryptographic hash of the exact model binary that produced the inference. This creates an unbreakable link between the decision and the specific model that made it.

Why Model Versioning Matters

AI models are updated frequently. A model deployed on Monday may behave differently than the version deployed on Friday. When a regulator, auditor, or attorney asks which model made a specific decision, organizations need a definitive answer. The MID provides exactly that: cryptographic proof of which model version was active for each decision.

How It Works

When a model is deployed, InferTrust™ computes a SHA-256 hash of the model binary (weights, architecture,and configuration). This hash becomes the MID for all decisions made by that model version. The MID is included in the signed decision record, making it tamper-evident alongside all other decision data.

Audit Scenarios

• Regulatory inquiry: prove that a specific, validated model version made the decisions during a given time period
• Incident investigation: identify exactly which model version was responsible for a specific decision
• Model rollback: demonstrate that a problematic model version was replaced and that subsequent decisions used the corrected version
• Compliance reporting: provide evidence that only authorized, validated model versions were used in production