AI in medicine is the use of artificial intelligence (AI) to support patient care. Each algorithm reads medical images, health records or clinical notes and helps clinicians detect disease, document visits and prioritize work. It does not replace the doctor. When a tool informs diagnosis or treatment, the FDA regulates it as a medical device, and in Europe it also falls under the EU AI Act.
What you’ll take from this
This guide is for health professionals, practice managers and anyone who wants a clear view of AI in healthcare without the hype.
- Tell apart the types of AI and machine learning used in the field of medicine.
- See where clinicians already use these tools, with official FDA figures from 2026.
- Understand the risks, the ethics and what US and EU rules require.
- Use a five-step checklist before you try a new tool in your practice.
What is medical AI? Examples of AI in health care
Medical AI covers AI algorithms that learn from medical data to handle tasks that used to need a person’s judgment. They read a scan, summarize a medical record, flag a risk or draft a report, and they return a probability, an alert or a draft that a clinician has to review.
A model does not reason like a medical doctor. It finds patterns in large datasets, and it can help predict disease risk or spot a finding early. Learning to use these tools with clinical judgment, and not only knowing they exist, is the focus of our Spanish-language AI in Health program at Founderz.
What types of AI are used in medicine?
Three families of technology, all rooted in data science, do most of the work in clinical care and practice. Machine learning learns to classify from labeled examples, such as thousands of ECGs, and deep learning, built on neural networks with many layers, drives the computer vision behind medical image analysis, from X-ray to magnetic resonance imaging (MRI).
Language models, the generative AI behind chatbots, work with text: clinical notes, guidelines and patient questions. All of these AI models depend on large, well-labeled data. Our guide to the types of AI places each technique on the wider map.
Applications of AI: how is AI used in health care?
The use of AI clusters around five areas: imaging, clinical documentation, risk alerts, medical research and health system operations. In every one, the system suggests and the clinician decides. The table shows what each AI-powered tool does and who keeps responsibility.
| Area | What the tool does | Example | Who decides |
|---|---|---|---|
| Medical imaging | Flags findings on X-rays, CT scans, mammograms or ultrasound | Moving studies with a possible bleed or tumor up the radiologist’s list | The specialist who signs the report |
| Clinical documentation | Uses speech recognition to transcribe the visit and draft the note | Ambient scribes that write a draft while you talk to the patient | The physician, who reviews and signs |
| Predictive alerts | Works as clinical decision support and scores the risk of deterioration from heart rate, blood pressure and lab results | Early sepsis warnings on acute care wards | The care team that responds |
| Research | Runs analytics on clinical data and peer-reviewed literature | Screening drug candidates or matching patients to trials | The principal investigator and the ethics board |
| Operations | Streamlines admin workflows | Scheduling, reminders and routine patient messages | Practice or hospital management |
Medical imaging, diagnosis support and surgery
Imaging is where the technology is most mature. According to the Stanford HAI AI Index 2026, radiology accounts for 76.6% of the AI-enabled devices the FDA had authorized through December 2025. Every other specialty, robotic surgery included, shares the remaining 23.4%. When a smart feature in a surgical system informs clinical decisions, it is regulated as part of the device, and the surgeon stays in control.
Drug discovery and personalized medicine
In this area, AI-driven tools help screen compounds, design trials and analyze clinical data to tailor therapy. The FDA reports that its drug center has seen more than 500 submissions with AI components from 2016 to 2023, and in January 2025 it published draft guidance on using AI to support regulatory decisions for drugs and biologics.
How to streamline operations and documentation
Operations are the safest place to start with digital health tools, because they never touch a diagnosis. Drafting notes into electronic health records, summarizing a patient’s medical history before a visit and sorting referrals all ease the administrative burden on providers, and in daily clinical practice the clinician still checks every note before signing it. If you run a clinic, our guide to AI tools for medical centers covers appointments and patient communication in detail.
How many AI devices has the FDA authorized?
The FDA’s official list counts more than 1,600 AI-enabled devices authorized as of September 2026. The FDA page is explicit that it does not regulate AI as such. It regulates devices, and software can count as one.
The Stanford HAI report adds the number worth remembering. Among devices authorized through December 2024 that had clinical studies, only 2.4% were backed by randomized trial data. Authorization does not prove a tool will help patients like yours, so ask for the evidence and the population it was validated on.
Published results from hospitals
The same Stanford HAI report collects measured cases. Northwestern Medicine cut time spent on notes by 24% with a documentation assistant. At UC San Diego, the COMPOSER sepsis alert was associated with a 17% reduction in sepsis mortality across 6,217 admissions.
These results come from specific hospitals and do not transfer automatically to yours. Our guide to artificial intelligence in hospitals walks through adoption department by department.
Is there medical AI for patients?
Yes, but it is not built for diagnosis. OpenAI launched ChatGPT Health on January 7, 2026 as a separate space in ChatGPT for understanding test results and preparing for appointments, and OpenAI states that it is not intended for diagnosis or treatment.
According to OpenAI, a company-reported figure, more than 230 million people ask health and wellness questions on ChatGPT every week. The company says conversations in Health are not used to train its foundation models. Its July 2026 update opened Health to US users 18 and older on web and iOS. Use it to organize symptoms and questions, never to change a medication.
What are the benefits and risks of medical AI?
The main benefit is time: AI can be used to help clinicians document faster and prioritize urgent studies, which leaves more of the visit for the patient. The impact of AI also brings known risks, and the World Health Organization lists them in its January 2024 guidance on large multi-modal models in health.
- False, inaccurate, biased or incomplete answers that can lead to a poor health decision.
- Poor-quality or biased training data, skewed by race, sex or age, so the model performs worse for some patients.
- Automation bias, when clinicians or patients miss an error or hand a hard decision to the model.
- Cybersecurity attacks that compromise patient data or the reliability of care.
- Unequal access, because the top-performing models are out of reach for many health systems.
Challenges and ethics
Ethics here comes down to who stays accountable and who could be harmed. The WHO set out six guiding principles in 2021, including protection of human autonomy, transparency and explainability, and inclusiveness and equity. Black box models make that principle hard to meet. Explainable AI, which shows why a model reached its output, helps human experts check it, and developing AI with that in mind is part of the answer.
Will AI replace doctors? How does AI work with the clinician?
No. Healthcare AI gives clinicians more information to diagnose and treat, but clinical decisions and accountability stay with the health professional who signs off. In practice, the system works like a second reader: it double-checks a scan or a note and the doctor makes the call.
The WHO describes uses such as diagnosis support, admin automation and medical education with simulated patients, always under human oversight, which is why training in AI matters for every clinician. A language model can invent a reference with full confidence. Professionals who know how to check the medical information they get back catch those errors first.
What does the law say about artificial intelligence in healthcare?
In the US, software intended for diagnosis or treatment is a medical device under FDA rules, and health information falls under HIPAA. In the European Union, the same software must meet the Medical Device Regulation and, in most clinical cases, the EU AI Act.
United States: FDA and HIPAA
The FDA reviews these devices through its usual pathways and publishes the list of authorized tools. Under HIPAA, a vendor that handles protected health information for a clinic or hospital needs a business associate agreement, so check for one before any records reach a new tool.
European Union: MDR and the AI Act
Software that informs diagnostic or therapeutic decisions is at least class IIa under the Medical Device Regulation, and needs a notified body: an independent organization designated by an EU country to check that a product meets the law before it goes on sale. The EU AI Act (Regulation 2024/1689) treats those products as high-risk.
Those high-risk obligations apply from the date set by Regulation (EU) 2026/1744, the Digital Omnibus on AI, which pushed back the original timetable. Medical device rules were not postponed. Under the GDPR, health data is a special category: data privacy rules require a clear legal basis and stronger safeguards.
What is the future of AI in medicine?
The near future already has dates and numbers. The FDA list grew from 1,357 devices at the end of 2025, as counted by Stanford HAI, to more than 1,600 by September 2026. That pace shows a growing role across the healthcare industry and the future of medicine.
In Europe, the high-risk rules of the AI Act and the European Health Data Space will shape what clinicians can use. The data space, in force since March 26, 2025, starts cross-border exchange of patient summaries and e-prescriptions in March 2029.
How to start using AI in your practice, ethically
The safest way for healthcare providers to start is with an admin task, data that cannot identify a patient and a person who reviews every output. These five steps work for a solo practice and for a hospital department.
- Pick a task with low clinical risk and high time cost: reminders, after-visit instructions or guideline summaries.
- Check what the tool is. If it claims to support diagnosis, ask for its FDA authorization or CE marking and the clinical evidence behind it.
- Review the privacy terms: where data is stored, whether it trains models and whether a business associate agreement is in place.
- Test it with de-identified data for two weeks and compare the output with what you would have done.
- Write down who reviews, what can be automated and how patients are told.
Here is a step 1 task you can try today in Copilot or ChatGPT, with no patient details.
Act as a primary care nurse. Write home care instructions after a tooth extraction for an adult with no other conditions. Use short sentences, plain language and no more than 150 words. Include when to call the clinic. Do not invent medication doses: leave a blank for the clinician to fill in.
Always review the text before you hand it over, especially anything about medication.
Frequently asked questions
These are the questions people search for most, with short answers.
Which AI is best in medicine?
No single tool fits every task, because each one solves a specific problem. A tool that reads mammograms will not document a visit. Choose by task, check FDA authorization or CE marking when the tool touches clinical decisions, and ask for studies on patients similar to yours.
Is Gemini or ChatGPT better for medical questions?
Neither should be your diagnostic tool. OpenAI states that ChatGPT Health is not intended for diagnosis or treatment, and a general chatbot is not an FDA-authorized diagnostic device. Use either one to prepare questions for your doctor, and check any answer against your clinician or an authorized tool.
Can AI prescribe medicine?
No. It can draft a note or flag a possible interaction, but a licensed prescriber has to decide on and sign any prescription. Consumer chatbots are not intended for diagnosis or treatment.
Is patient data safe with these tools?
It depends on the vendor and the contract, not on the technology itself. Before you use a tool, check where data is stored, whether it is used to train models and whether a business associate agreement is in place. With identifiable records, get all three answers in writing.
Further reading
These official sources go deeper into the rules and data covered in this guide.
- Artificial Intelligence-Enabled Medical Devices, US Food and Drug Administration
- Ethics and governance of artificial intelligence for health, World Health Organization
- Regulation (EU) 2024/1689, the EU AI Act, EUR-Lex
If you work in health care and read Spanish, our Programa Online de IA en la Salud (taught in Spanish) covers AI-assisted diagnosis, health management, ethics and privacy with experts from Hospital Universitario La Paz and Microsoft. It is fully online and ends with a Founderz and Microsoft certificate, with official Microsoft badges along the way.
This article is for informational and educational purposes only. It is not medical advice and does not replace the judgment of a qualified health professional, so talk to your doctor about any health question. Use the tools mentioned here in line with the regulations that apply to you.
