Healthcare AI evidence

Signed records for healthcare AI workflows.

One metered evidence layer for the health system, AI vendor, broker, carrier, and reviewer. The issuer pays for each signed receipt. Anyone can verify it without exposing a chart.

HEALTH SYSTEMProduce a defensible record at audit, renewal, or incident review.
AI VENDORShow how the deployed product handled disclosure, custody, retention, and change.
CARRIER OR BROKERAsk for machine-checkable evidence instead of another questionnaire.

Six receipt types and the questions they address

Each receipt makes one narrow claim. A verifier checks the signature, ordering, links, and fixed fields. A reviewer can accept or reject the claim without seeing the underlying patient record.

01

Disclosure presentation

Was the recorded notice shown before the linked AI action?

02

Custody presentation

Did one party sign the proposed artifact, purpose, and retention terms?

03

Custody handoff

Did two different keys agree on the same terms and the order of transfer?

04

Retention policy

Was the policy version, scope, and digest fixed before it took effect?

05

Retention purge

Was a deletion event recorded against the earlier policy and erasure receipt?

06

Model change

Does the observed serving configuration match the one that was evaluated?

Commercial model: paid issuance by receipt, free public verification. Enterprise commitments purchase receipt volume and reconcile to actual use. There is no seat-based subscription.

Follow the record through four stages.

The same receipt package can move with the system through ordinary governance and the moments when evidence suddenly matters.

Deployment

Fix the policy, disclosure, custody terms, and evaluated model configuration.

Renewal

Give the broker or carrier aggregate evidence that the controls kept running.

Incident

Reconstruct what was presented, transferred, retained, purged, or changed.

Audit or claim

Let the other side verify each signed claim independently.

Inspect an example receipt.

Choose a public example. This page sends the signed object to the production Hive verifier and prints the result returned by the live service.

LIVE PRODUCTION VERIFIERREADY
Select a receipt to run the proof.

These are signed public examples. They contain no patient information. A valid result proves only the narrow claim named by the receipt.

Use alongside model validation.

Epic's Seismometer asks whether a model performs and is fair across local populations. Hive records whether the deployed system followed the disclosure, custody, retention, purge, and evaluated model-change rules around that model.

Performance and fairness

Use Seismometer or another validation system to measure accuracy, fairness, drift, interventions, and outcomes.

Behavior and evidence

Use Hive to give an outside reviewer a signed record of what the deployment did and when it did it.

Hive is not affiliated with or endorsed by Epic. The comparison describes complementary functions.

What it proves

  • Signed records and ordering
  • Links between policy, action, and evidence
  • Agreement or divergence between parties
  • Serving configuration match or difference
  • Independent verification without raw PHI

What it does not claim

  • A diagnosis or clinical recommendation
  • Patient comprehension or legal consent
  • Policy adequacy or regulatory compliance
  • A risk score, premium, or coverage decision
  • That every possible copy of data was erased
Design partner program

Evaluate one workflow over 90 days.

We will map one live healthcare AI workflow, issue the relevant receipts, run an evidence drill, and shape the export around the questions your legal, compliance, risk, or underwriting team already asks.