Alpha: working software, ready to be put to work. The demos on this site use sample data and demo keys for an invented company.

Use case · Healthcare

"It runs securely" is a promise. A receipt is evidence.

AI scribes listen to visits. AI helps decide prior authorisations. Vendors say it all runs in a secure environment. The people affected have no way to check.

What is happening

Three stories from this year.

  • Visits recorded by an AI scribe. Patients filed a class action saying their doctor visits were recorded and processed by a third-party AI system without clear notice. (reported by TechTarget, April 2026)
  • AI deciding prior authorisations. Medicare is piloting vendor AI on prior-authorisation requests. The Electronic Frontier Foundation obtained records and wrote: "Little is known about the AI systems that WISeR vendors are using to process prior authorization requests." (EFF, September 2026)
  • Evidence instead of assumptions. The confidential-computing industry itself now argues that a regulator should be able to verify the approved algorithm is the one actually running. (Confidential Computing Summit 2026 talk)

The question

What patients and reviewers can't answer.

"Which system processed my visit, under whose approval, and did the approved program actually run on my data?" Today the answer is a contract between the hospital and the vendor, which the patient never sees.

What a receipt adds

The result says how it was made.

  • Who approved the run: the responsible officer's signed policy, as of that moment.
  • The controls in force: where the data was allowed to go, as the sandbox enforced it.
  • Where it ran, stated honestly: the receipt says whether the computer was measured by hardware or only declared. With confidential computing, where even the operator can't see the data, this would become a verified line on the same receipt (planned).
  • Bound to the note or decision: change one byte and the check fails.

Honest limits

What it does not do.

A receipt does not prove a patient consented, and it does not prove a clinical decision was right. It proves which process produced the note or decision and under what rules, so consent and correctness can be argued about with evidence instead of promises.

Where we are: Rootz Receipts is in alpha. It runs today on NVIDIA OpenShell, with a live demo and a bank-collection pattern working against a test bank. Confidential computing is the planned upgrade; the compute behind today's demo is declared (a software TPM), not hardware-measured, and every receipt says so.

Try it on one workflow · See a receipt checked · Other examples