A CSR program from Diagnostics AI, Inc.

Premedice for the
Damned.

Free AI medical credits for people with serious conditions who have run out of conventional options.

Apply for credits How it works

100–10,000

USD in free credits per case

48 hrs

Typical application review time

Dollars ever required from an applicant

In one paragraph

Premedice for the Damned is a corporate social responsibility program operated by Diagnostics AI, Inc. (the maker of Premedice) that grants free Premedice-1 (the production medical expert model that scored 86.5% on the MedQA USMLE benchmark) and Premedice Gamma (an experimental beta with stricter refusal and longer reasoning) usage credits ranging from $100 to $10,000 per approved case, to adults and dependents with serious, ongoing medical conditions who have already consulted at least one qualified clinician and feel they have exhausted conventional options. Applications are reviewed by a human within 48 hours. The program is global, free, and has no income or insurance requirements. Premedice for the Damned is a second or third opinion, not a replacement for the medical system; every applicant must agree to verify any AI output with a clinician before acting on it.

The premise

We built the tool. Some people still cannot reach it. That is the problem this program exists to solve.

Premedice is a paid product. The cheapest plan is still out of reach for the people who would benefit from it most — the patients who have already spent on specialists, on travel, on imaging, on the wrong prescriptions, and on the long tail of appointments that did not move the needle. By the time someone finds us, they are usually not looking for another bill.

The Premedice for the Damned program exists for that patient. It is a CSR line, not a marketing line. It is not a free trial, not a discount, not a referral scheme. It is a small number of medical AI models we already operate, granted in credits to people who can show us, in a short email, that the conventional medical system has not yet given them an answer they can use.

The name is blunt on purpose. These are the cases medicine has not closed — the eleven-year diagnostic delay, the rare disease no local specialist has seen, the symptom cluster that does not fit any one specialty. We are not promising to fix them. We are promising to give one more tool to the patient and the clinician who are still working the problem.

What you get

Two models, a workspace, and a person on the other end.

The production Medical Expert. 86.5% on MedQA.

Premedice-1

The same model that powers the paid product. Reads your full chart, searches across 30M+ papers, drafts differential diagnoses in plain English, and labels every output with what it is certain about and what it is guessing. This is the model that does most of the work in approved cases.

The experimental beta. Stricter refusal, longer reasoning.

Premedice Gamma

An in-house experimental model with a longer reasoning chain and a more conservative refusal policy. It will sometimes refuse to answer where Premedice-1 will; that is on purpose. Useful as a second pass on cases where the first model sounds too confident.

Encrypted. Zero retention. AES-256 at rest, TLS 1.3 in transit.

A private workspace, pre-loaded

Your prompts, uploads, and outputs are not used to train any model. The same architecture that protects paid customers protects you. We do not sell applicant data, and we do not share it with any third party.

Two people review every application. One of them stays with you.

A direct line to a human

If the tool misbehaves, you can email a person. If you need a clinician review of an AI output, the program coordinator can route you to a partner telemedicine provider at a reduced rate (paid by you, not by us).

For background reading, the architecture of the model is described in the open zero-knowledge privacy breakdown, and the 2026 medical AI model comparison explains where Premedice-1 and Premedice Gamma sit among the published benchmarks.

What Premedice can do

Things a tired, well-read machine does that an overworked human often cannot.

This is not a sales pitch. It is an honest list of the work the model is built for, with the failure modes the program has actually seen. None of these are guarantees. Every one of them requires a clinician in the loop before any decision.

Capability 01

Reading the entire literature on a single rare disease in seconds.

Premedice-1 searches across more than 30 million indexed papers through PubMed, Europe PMC, ClinicalTrials.gov, the Cochrane Library, and several specialty databases. A clinician working a typical caseload reads a few hundred papers a year; the model reads them all, every session, and surfaces the ones that look like your case. That is not a small advantage — it is the difference between a textbook from 2014 and a textbook from last week.

Capability 02

Pattern matching across specialties.

Most specialists see cases inside their own field. Premedice sees across cardiology, rheumatology, neurology, and endocrinology in a single pass. Symptoms that look like three different diseases to three different specialists can read like one disease to a model trained on all three.

Capability 03

Generating new hypotheses, not just retrieving old ones.

This is the part no one markets. Large medical models are not just retrievers. They interpolate — they combine patterns from adjacent cases and produce a hypothesis that no published paper has yet described. Sometimes the hypothesis is wrong. Sometimes the hypothesis is a paper a researcher has not yet written. We surface those flagged in the model output so a clinician can decide which deserve follow-up.

Capability 04

Sitting with you while you wait.

Specialists are not available at 3 a.m. The model is. Most of what a desperate patient needs at 3 a.m. is not a diagnosis — it is a translation of what they already have. Premedice is well-tuned for the long, anxious, midnight question.

Important: model can interpolate, not just retrieve.

The model is not a search engine. It can interpolate between adjacent cases and produce a hypothesis that no published paper has yet described. That is a feature of the architecture, and it is the part most likely to find something no one has named. It is also the part most likely to be wrong. Every AI-generated hypothesis in the program output is labelled, and a clinician reviewer on the program has final say on which deserve follow-up. If you act on an AI output without a clinician review, that is on you, not on us.

What Premedice cannot do

Read this section before you apply.

01Premedice cannot examine you. It cannot feel a lymph node, listen to your lungs, take your blood pressure, or look into your eyes. Anything that needs an embodied exam is not in scope.
02Premedice can hallucinate. It will sometimes invent a citation, misattribute a finding, or sound confident about something it is not. We surface the model uncertainty in the output. Always verify.
03Premedice is not for emergencies. If you are in immediate danger, call your local emergency number first. The program is a second or third opinion, not a substitute for the medical system.
04Premedice does not replace your doctor. The model can read your chart; your doctor knows your life. The program works best when the model and the clinician work together.

Premedice for the Damned is a second or third opinion for people who still have a clinician in their life — or who are willing to find one. It is not, and will never be, a replacement for the medical system. If you do not have a clinician, we can connect you with one of our partner telemedicine providers at a reduced rate (paid by you, not by us). If you are in immediate danger, call your local emergency number first.

How it works

Three steps. A human reads every application.

Step 01

Apply by email

Click the button below. Your email client opens with a short form pre-filled. Tell us your situation, what you have already tried, and why you think Premedice might help.

Step 02

We review in 48 hours

A human reviewer (one of two people on the program) reads every application. We either approve with a credit grant, ask for a short clarification, or decline with the reason. No automated screening.

Step 03

You get a private workspace and credits

Approved applicants receive a Premedice workspace with the credit grant loaded, access to Premedice-1 and Premedice Gamma, and a direct line to the program coordinator for any technical or clinical question.

Eligibility

You can apply if all five of these are true.

01You have a serious, ongoing medical condition.
02You have already consulted at least one qualified clinician in person or via telemedicine.
03You have reliable internet access and the ability to use a web tool.
04You understand Premedice is not a doctor and agree to verify any output with a clinician before acting on it.
05You are applying for yourself, or for a dependent with documented consent.

Three anonymized cases

What the program has actually done.

Every case below is real and anonymized with the patient's written consent. We publish the failures alongside the wins because the program only matters if you can tell the difference.

Case 01 — Anonymized

An eleven-year diagnostic delay, ended in a single session.

A patient in their thirties had spent eleven years moving between specialists for an unexplained constellation of symptoms — fatigue, intermittent neuropathy, atypical migraines. Premedice-1 was given the full chart and the timeline. In a single session it surfaced a 2019 case series on a rare mitochondrial disorder that matched the symptom cluster and pointed at a confirmatory test the patient's specialists had never ordered. The test came back positive. The patient is now in treatment. Premedice did not diagnose the patient; the patient's neurologist confirmed the result. The model did what models can do well: it read widely, and it did not get tired.

Case 02 — Anonymized

When the model was wrong, and what we learned.

An applicant with a long-standing autoimmune presentation uploaded imaging and labs and asked for a differential. Premedice-1 confidently suggested a diagnosis the applicant had already been excluded from by a prior biopsy. We walked the model output back through the case with a clinician reviewer. The model had weighted a single lab value out of proportion to the full picture. We use that case now as a training example inside the program: any Premedice output that contradicts a biopsy, an imaging read, or another hard clinical finding should be flagged to your clinician before you change any plan. The model is a second pair of eyes, not a tiebreaker against tissue.

Case 03 — Anonymized

A specialist in a country that does not have one.

An applicant in a region with no practicing neurologist and a four-hour drive to the nearest internist had been managing an inconclusive imaging report on their own. Premedice-1 walked them through the report line by line, surfaced the three items that needed follow-up, and drafted the exact language to bring to the next available generalist. The generalist confirmed the plan. The applicant did not need a specialist to read the report; they needed a tool that could help them ask the right questions of the clinician they did have. That is the work this program was built for.

Questions people ask before they apply

Frequently asked questions

Q01

Who can apply?

Anyone with a serious, ongoing medical condition who has already consulted at least one qualified clinician and feels they have run out of conventional options. The program is global, but applicants must have reliable internet access to use the tool. We do not require proof of income; we trust applicants to be honest about whether they could afford a paid plan.

Q02

How much does it cost?

Nothing. Premedice for the Damned credits are free, no money changes hands, and there is no obligation to pay anything back. We do not ask for insurance, financial documents, or a credit card. The credits never expire.

Q03

How are credit amounts decided?

Credit grants range from $100 to $10,000 in Premedice usage, sized to the case complexity and the expected volume of model calls. A straightforward case that needs a few deep reviews may receive a smaller grant; a case that needs many sessions across multiple specialists can receive the upper end. The grant is set in the approval email, not on the application form.

Q04

How long does approval take?

Most applications are reviewed within 48 hours during European business days. Edge cases (active emergencies, complex psychiatric situations, minors) are reviewed by a clinician reviewer and may take up to 5 business days. We do not process applications over weekends or European public holidays.

Q05

What models do I get?

Approved applicants get access to Premedice-1, the production Medical Expert model that scored 86.5% on the MedQA (USMLE) benchmark, and to Premedice Gamma, an experimental beta that uses a longer reasoning chain and a stricter refusal policy. Both are routed through the same encrypted pipeline as the paid product, with the same privacy guarantees.

Q06

What does Premedice actually do that a doctor might miss?

Premedice is best at tasks with very large evidence bases and at pattern matching across specialties. It can read every paper ever indexed on a rare disease in seconds, and it can surface differential diagnoses a single specialist might not consider because they live outside that specialist's training. It is less useful for embodied clinical judgment, surgical planning, and the relational side of medicine. The model can also generate hypotheses (the literature sometimes calls this 'interpolation') that no published paper has yet described — that is a feature, but it must be verified, not a diagnosis.

Q07

Is this safe?

Premedice is a clinical information tool, not a clinician. It can and does make mistakes. We require every applicant to confirm they will verify any AI-generated output with a qualified clinician before acting on it. Premedice for the Damned is not for emergencies — if you or someone you care for is in immediate danger, call your local emergency number first. The program is a second or third opinion, not a replacement for the medical system.

Q08

What happens to my data?

All applicant communications are encrypted in transit and at rest (AES-256, TLS 1.3). The same zero-retention architecture used by the paid product applies: your prompts and uploads are not used to train any model, and we do not sell or share applicant data with any third party. The full privacy policy at /privacy applies.

Q09

Can I share my credits with someone else?

No. Credits are tied to the applicant account and the medical situation described in the application. If a family member or caregiver needs their own credits, they should apply separately. Sharing credits is grounds for revocation.

Q10

What if Premedice tells me something my doctors disagree with?

That is not unusual and is part of why the program exists. Bring the AI output to your clinician and ask them to evaluate it. If you do not have a clinician, we can connect you with one of our partner telemedicine providers at a reduced rate, but we never substitute the model for a human review.

Apply

If this is the program you have been looking for, the next step is a short email.

Click below. Your email client opens with a form pre-filled. Tell us the situation in your own words. A human reads it within 48 hours and writes back.

Apply for credits

Or write directly to [email protected]