How we keep an AI that reads your labs safe

A lab result is not just a number. It sits inside a person's history, their medications, their symptoms, and a clinician's judgment. So when the Northstar team asked me to help think through the safety layer of a product that reads lab panels and explains them, we spent most of our time on one question: where is the line between helping someone understand a result and telling them what it means for their health? Those are not the same thing, and the whole design depends on keeping them apart.

It explains. It does not diagnose.

Northstar can tell you that your LDL cholesterol is above the usual range, roughly by how much, and what LDL generally reflects. It will not tell you that you have a condition, what is causing the result, or what to do about it. That distinction is not a limitation we are apologizing for. It is the point. Diagnosis belongs to your clinician, who knows your full history and can examine you. Northstar knows the numbers you gave it. Explaining those clearly is genuinely useful, and it is also genuinely not medical advice.

It never gives medication or dosing guidance

If you ask Northstar whether to start a supplement because your vitamin D is low, or whether to change a dose because a number moved, it will not answer. It will say plainly that it cannot help with that and point you to your clinician or pharmacist. The same holds for anything framed as treatment. This rule is enforced in the model's instructions and again in code, so it holds even on an unusual result.

• No medication, dosing, or supplement guidance, ever.
• No naming a diagnosis, disease, or cause as fact.
• No treatment or dietary instructions.
• No reassurance that a result is fine when a clinician should weigh in.

What happens when a value or a note looks urgent

Two things trigger a clear route-to-care note. If a value comes back far outside the usual range, Northstar flags it as worth a prompt conversation rather than quietly filing it away. And if a note you write mentions a red flag, such as chest pain, trouble breathing, fainting, or one-sided weakness, Northstar does not try to assess it. It surfaces the information and tells you directly to seek care now. It does not diagnose, and it does not minimize. It points toward a person, quickly.

The most important thing a health AI can do is know when it is not the right tool for the moment, and say so plainly.

Reference ranges are a guide, not a verdict

Northstar uses general adult reference ranges to help you read a result, and it says so every time. Ranges vary between labs, between methods, and between people. A value just outside a general range is not automatically a problem, and a value inside it is not automatically fine for you. Northstar is careful to frame ranges as context for a conversation, not as a pass or fail.

Your data is yours

Northstar does not sell your health data and does not use it to train commercial models. You can export everything you have added, at any time, and you can delete it. Trust in a health tool has to be earned slowly and kept carefully. Northstar is not trying to replace the people who care for you. It is trying to make your own numbers make sense, and being honest about its limits is what makes the rest of it work.