Prediction, Risk and Biological System Models for Public Health, Human Protection and Drug Safety, designed against PAVETS, the six requirements for AI systems in healthcare.
PAVETS is the framework Health Enovation uses with clients to judge whether an AI system in healthcare is ready for inspection. Each model is first given a risk tier (Low, Medium or High) from its context of use, based on how much the model influences a decision and how serious that decision is. It is then held to the bar for that tier on six dimensions.
| Requirement | How Tellura designs for it | |
|---|---|---|
| P | Privacy Data protection and permitted use across the training, tuning and inference data. | Consent is captured separately from qualification, and research licensing is opt-in. Only consent records and payouts go on the ledger, never health data. |
| A | Accountability Named owners, human oversight design and escalation paths. | Outputs are decision support for flight medical teams and safety reviewers. A qualified person makes every decision. |
| V | Validation Fit-for-context-of-use performance evidence, drift monitoring and change control. | Each model is scoped to a defined context of use. SilicoSafe is a research prototype, and its validation is in progress. |
| E | Explainability Whether intended users can understand outputs, limits and uncertainty. | Every medication flag shows the evidence behind it and a suggested action, so users can see why the model flagged it. |
| T | Traceability Audit trails and data and model lineage from source to output. | Outputs are designed to trace back to their public sources (FDA, NASA and FAERS data) and to the model version that produced them. |
| S | Security Protection of models, pipelines and outputs against tampering. | Security is designed into the platform architecture, from the data pipelines to the models and their outputs. |
How we design for these requirements is not a certification or a regulatory finding. No regulator has reviewed or endorsed these models. Validation is in progress.