Asuu

In plain language

The whitepaper, in plain language.

A hundred and eighty-seven pages, read for you in ten minutes.

A plain-language reading of a published proposal. Where the two differ, the paper is the authority.

On this page

The idea in one sentence.

The whitepaper proposes a network that sits beneath medical AI and records which model acted, who allowed it, and how it changed, in a way that no single company or government controls.

Where the data stays.

The paper calls its central idea the data embassy. A hospital's patient records stay inside the hospital, under that country's law, as an embassy stays under its own flag. Only fingerprints and mathematical proofs leave the building.

A zero-knowledge proof can establish a specified statement while concealing the witness used to establish it. The proposal explores this for eligibility checks. What remains private depends on the statement, public inputs, metadata, and the way the system is implemented.

The first workflow.

The paper picks one unglamorous task to begin with: the request a clinic sends to an insurer before a treatment is approved. It is slow, heavy with paperwork, and disliked on every side.

This section is unusually frank. It lists the reasons an insurer might adopt such a system, and then the reasons it would refuse. An insurer will not give up its right to decide, and will not accept a record that holds it to account while leaving clinics unexamined. So the design proposed is a hybrid. The system checks the documented criteria and records the result. The insurer keeps the decision, except in the kinds of case it has agreed beforehand to hand over.

A passport for every model.

In place of a static description of a model, the paper proposes a passport that travels with it and is updated at each change. It states what the model is for, how it has performed, and which changes are permitted without fresh approval.

Regulators and the model's sponsor keep a separate channel through which they can order a model suspended. The institutions keeping the record cannot widen what a model is permitted to do on their own.

The human stays in charge.

The paper sets out to make human oversight a property of the system and not a line in a policy. An action that needs a clinician's authorisation cannot be recorded as complete without it.

The proposal defines permissions by role in seven tiers, combining credentials, authorisation rules, and cryptographic mechanisms. It also specifies an emergency route for access outside usual permissions, with a record of its use. These controls still need implementation and evaluation.

Who answers when it goes wrong.

When several parties contribute to a harm, the paper proposes dividing responsibility using a method from game theory, applied to the recorded trail. It then spends a section on the cases where that method fails, such as when several parts of a system break for the same reason.

Where it reaches further.

Later chapters carry the same record into other fields: following a medicine through its supply chain, keeping the conduct of a clinical trial honest, noticing the side effects of a drug across many hospitals, and preparing for a pandemic.

One chapter joins disease surveillance in people, in animals, and in the environment. The findings are currently reported to separate bodies. Another covers a clinician’s qualifications, so that they can be checked in another country. Readers of this site will recognise where several of our proposed ventures began.

A hospital's records stay inside the hospital. Only fingerprints leave the building.

What is still only on paper.

All of it. There is no live network. The paper includes an economic design in which the institutions keeping the record put something at stake. None of that operates today. The figures given for a pilot are targets. The formal checks of the proofs are planned and have not been published.

The whitepaper is a proposal, set out so that people who know more than we do can find its faults. The next step is the small evaluation described elsewhere on this site, and the paper should be judged again when that is done.

Where this connects.

Good questions move the work forward.

Share one workflow problem, one protocol objection, or one evaluation requirement. Your email goes to Berkcan to begin a conversation.

Sources and scope

This summary describes the published proposal under its stated assumptions. A proof about a model does not establish clinical safety or the security of an implementation.

Original paper in English