Our principles

Four principles — and the hard parts of holding to them.

The homepage states what AuVentures believes. This page does the harder thing: it explains why each principle matters, and where it is genuinely difficult to uphold. An honest account of our ethics has to include both.

Principle one

Patient information serves the care of the patient it belongs to.

It has become common to say a patient “owns” or “controls” their data. We say it too. But the phrase deserves more honesty than it usually gets — because in the digital world, and even more so with AI, very few people actually know what that control amounts to.

Consider a familiar example. A patient uploads their records into a free AI chat tool and finds it genuinely helpful. It feels like control: they chose the file, they chose to upload it. But free consumer AI tools are often supported by advertising or data-monetization models, and uploading a medical record into one can open downstream uses the person never sees and never agreed to. They controlled the upload. They did not control what happened next.

Controlling the moment of upload is not the same as controlling what happens behind the scenes, or downstream, or later.

So when AuVentures says patient information serves the care of the patient it belongs to, we mean it as a constraint on the whole lifecycle of that data — not just the moment it enters the system. Information is used for the patient’s care, and not quietly repurposed. If research could be done with a patient’s data, we will tell them, ask whether they wish to take part, and — where they do — offer direct connections to the providers and researchers involved. Participation is a decision the patient makes, not a default they are enrolled into.

Real control requires real transparency. We believe a patient cannot meaningfully control what they cannot see — so the work of this principle is as much about showing people what happens to their information as it is about giving them switches to set.

Related writing from our research is forthcoming in Notes & Essays.

Principle two

Clinicians remain accountable for clinical decisions.

The straightforward reading of this principle is that AI should not make medical decisions. That is true, and we hold to it. But there is a harder problem underneath it — one that is less about the technology and more about the people we need at the table.

Clinicians have not been well served by the technology placed in front of them. Many of the systems they use every day — electronic health records chief among them — were not designed for their efficiency or for patient outcomes. They were shaped to maximize insurance claims, to operationalize hospital processes, and to provide legal protection when something goes wrong. The work of genuinely understanding a patient gets squeezed into whatever time is left over. Dr. Ilana Yurkiewicz’s book Fragmented describes the lengths to which doctors go to care for patients inside systems not built to support them; it is a book that continues to influence how we think about this work.

Given that history, a clinician’s wariness toward yet another piece of software is not a failing. It is a reasonable, learned response. But it creates a real risk — and the risk is not what people usually assume.

The danger is not that clinicians will lean on AI too heavily. It is that they will, understandably, opt out of it — and healthcare AI will then be built without them.

If the people who actually deliver care step back from these tools, the future of AI in medicine gets shaped without clinical judgment in the room. The ethics, the design, the guardrails — all developed in the absence of the practitioners who understand what care actually requires. That is the outcome this principle is meant to guard against. Keeping clinicians accountable is not only about the present decision in front of a patient; it is about keeping clinical wisdom inside the development of the technology itself.

So earning a skeptical clinician’s time is not, for us, a marketing problem. It is part of the ethical work.

Related writing from our research is forthcoming in Notes & Essays.

Principle three

People should be able to understand how AI is being used.

This principle is easy to state and genuinely difficult to honor — because it depends on something larger than any single product. It depends on people understanding AI well enough to think clearly about it.

There are a great many misconceptions about what AI is, what it does, and what it cannot do. And the technology is evolving faster than the ethical guidelines meant to govern it. Rules written for one generation of systems are often outdated by the next. In that environment, “trust our ethics” is not a sufficient answer — because ethical frameworks themselves are still catching up.

When the rules can’t keep pace, people need to understand AI well enough to apply their own ethical judgment to it.

That is the deeper purpose of this principle. We commit to explaining how AI is used in a patient’s care plainly, without obscuring the details that matter — what systems are involved, what role they play, and why. But the goal is not only disclosure. It is to help patients and clinicians understand these tools well enough to bring their own values to bear on them — to ask their own questions and reach their own conclusions, rather than defer to ours.

Transparency, done well, is a form of respect: it treats the other person as someone capable of ethical judgment, and gives them what they need to exercise it.

Related writing from our research is forthcoming in Notes & Essays.

Principle four

We document what we get wrong.

AI in healthcare is cutting-edge work. It is not perfect, and it will make mistakes. Any organization claiming otherwise is being either inaccurate or willfully ignorant — and in this field, that posture is itself a kind of harm.

So we commit to the opposite. When the system gets something wrong, we will acknowledge it, examine it, and use it to improve the work responsibly. This is not damage control. It is a stance about how knowledge in this field should be built.

Trust is built by what an organization is willing to admit — not only by what it promises.

There is a harder edge to this principle, and we want to name it plainly: documenting mistakes openly can carry real cost. It is easy to commit to transparency in the abstract and harder to honor it when a particular disclosure is uncomfortable. We hold the commitment anyway, because the alternative — a healthcare AI field where mistakes are hidden — fails exactly the patients who can least afford it.

We also believe sharing what we learn serves a purpose beyond our own accountability. It advances research for underserved populations, it helps the broader community learn faster, and — perhaps most importantly — it lets people see what ethical AI actually looks like in practice, rather than only hearing it described.

Our open record of what we get wrong, and related writing, is forthcoming in Resources.

A note on these principles

These four principles are not finished. As AuVentures’s work develops — and as the research behind it continues — we expect our understanding of them to deepen, and we expect to find hard cases we have not yet anticipated. When we do, this page will change.

That, in a sense, is the point. A set of principles that never has to be revised is probably not being tested against reality.

Questions about our principles, ethics, or data practices: privacy@auventureshealth.org