Will AI replace doctors? What the data says in 2026

No, AI is not going to replace doctors, but it is already taking over a growing share of the paperwork and pattern-spotting around them. The U.S. Bureau of Labor Statistics (BLS) rates most physician specialties as highly exposed to AI, yet it still projects physician and surgeon jobs to grow 4% by 2035, with about 22,100 openings a year.

The reason is that most of a doctor’s value sits in the parts of the job AI can’t reach: examining a patient, making a call under uncertainty, doing a procedure and answering for the result. AI is getting good at the parts that look like reading and writing, such as scans, records and notes. Below are the official numbers, what AI can and can’t do in medicine, which health jobs are most and least at risk, and how to stay ahead of the change.

The numbers: exposed, but still growing

In August 2026 the BLS published official AI exposure ratings for 831 occupations, alongside its job projections for 2025 to 2035. The ratings combine five research measures, including real-world AI usage data from Anthropic and Microsoft, and sort each occupation into one of four roughly equal groups: low, moderate, high or very high exposure.

The BLS splits physicians into 17 specialties. None of them is in the very high group. Eleven, including family medicine, internal medicine, psychiatry and radiology, are rated high. Five surgical and procedural specialties, such as orthopedic surgeons and dermatologists, are moderate, and anesthesiologists are low. Here is how that compares with where the jobs are expected to go:

Every physician specialty is projected to grow. Overall, physician and surgeon jobs rise from 862,800 in 2025 to about 894,100 in 2035. The BLS ties that growth to an aging population that needs more complex care and to rising rates of chronic illness such as diabetes, cancer and heart disease. Its outlook for doctors doesn’t name AI as a reason for slower demand at all.

The Association of American Medical Colleges goes further: it projects a shortage of up to 86,000 physicians by 2036. A profession that is short of people is not one where software is about to make people surplus.

Pay reflects that position. The median physician or surgeon earned $275,930 in May 2025, more than five times the $50,980 median for all occupations. Radiologists, the specialty most often named as next in line for AI, earned a median $420,860.

Can AI replace doctors? What it can and can’t do

The most useful evidence comes from how AI is actually used and tested, not from predictions about what it might do one day.

Microsoft analyzed 200,000 anonymized conversations with its Copilot assistant and scored 785 occupations on how much of their work AI was used for, and how well. Family medicine physicians rank 334th and internal medicine physicians 341st, close to the middle. Radiologists rank 400th, emergency physicians 601st and anesthesiologists 728th. For comparison, the health jobs that rank highest are the information-heavy ones: genetic counselors (79th), health information technologists (87th), pharmacists (119th) and medical records specialists (131st).

Anthropic, which measured which work tasks its Claude models are really used for, makes a related point. It gives the example of a medical task that an AI model could in theory speed up, authorizing drug refills and passing prescription information to pharmacies, and notes that it had not seen Claude doing it in practice. Legal limits, software requirements and human checks keep many tasks like this in human hands even when the technology looks capable. No health occupation appears among the most exposed jobs it names.

Clinical trials show the same gap between capability and real care. In a 2024 randomized trial published in JAMA Network Open, 50 physicians, 26 attendings and 24 residents, worked through difficult diagnostic cases. Those given a chatbot as well as their usual resources scored a median 76%, against 74% for those with usual resources only, a difference too small to count. Yet the chatbot on its own scored 16 percentage points higher than the doctors without it. The tool knew a lot; the doctors didn’t get much better at using it. Fitting AI into how doctors actually think is the hard part.

Here is how a doctor’s work splits:

• Most exposed: documentation, records and first-pass reading. Writing visit notes, summarizing a chart, drafting letters and prior authorizations, flagging findings on an image. AI does a fast first version; a clinician must check it.
• Partly exposed: diagnosis and treatment planning. AI can suggest possibilities and check guidelines, but weighing them against an exam, a history and a patient’s wishes is still the doctor’s job.
• Least exposed: examining, procedures and hard conversations. Physical exams, surgery, emergencies, delivering bad news and agreeing on a plan with a frightened family. These rest on hands, judgment and trust.

AI is taking the paperwork first

The clearest early change is not diagnosis but documentation. Ambient AI scribes listen to a visit, with the patient’s consent, and draft the note. In a 2025 study across six U.S. health systems, published in JAMA Network Open, 263 clinicians used an AI scribe and reported burnout fell from 51.9% to 38.8%. They also reported less after-hours documentation. The study had no control group, so the effect may be smaller in practice, but the direction matches what doctors describe: AI gives back time rather than taking away the job.

That shift explains the one decline in the table. Medical transcriptionists, who turn dictated notes into records, are projected to lose 4.4% of their jobs by 2035. Of all the health occupations the BLS rates high or very high for AI exposure, it is the only one projected to shrink. The work AI replaces in medicine is the typing around care, not the care.

Will AI replace radiologists?

Radiology is where AI has gone furthest. Of the 1,614 AI-enabled medical devices on the U.S. Food and Drug Administration’s public list, 1,230, or 76%, are radiology devices. The FDA notes the list is not comprehensive, but no other specialty comes close: cardiovascular devices are a distant second, with 154.

If any doctors were going to be replaced first, it would be radiologists. Instead the BLS projects radiologist jobs to grow 3.4% by 2035, rates their exposure as high rather than very high, and Microsoft ranks them in the middle of its list. The tools speed up parts of reading an image. Deciding what a finding means for this patient, talking to the doctor who ordered the scan and doing image-guided procedures remain the radiologist’s work.

Why AI can’t legally replace a doctor

There is also a hard limit: practicing medicine requires a license, and software can’t hold one. Only a licensed clinician can diagnose, prescribe and take legal responsibility for a patient’s care.

States are writing that line into law. California’s AB 489, approved in October 2025, bars AI systems from using terms or phrases that suggest their advice, care or assessments come from a licensed health professional. Each misleading use counts as a separate violation, enforced by the relevant licensing board. AI can support a doctor; it can’t pose as one.

Which health jobs are most and least at risk?

The jobs most exposed to AI are not the clinical ones but the information jobs around them. In the BLS data, medical records specialists, health information technologists, medical secretaries and genetic counselors are all rated very high. All four are still projected to grow, because healthcare itself is growing, but their daily work will change the most. Medical transcription is already shrinking.

The least exposed are the hands-on roles. Physician assistants, nurse anesthetists, paramedics, surgical technologists and nursing assistants are all rated low. Our own scoring, which uses the same method as our free test on a typical week, shows the same pattern: jobs built on physical, in-person work score lowest of all.

A typical nurse scores 16 out of 100, close to a plumber or a chef, because most of the week is spent with patients. These three health careers in our index combine low exposure with strong demand:

Physician assistants stand out: rated low for AI exposure, projected to grow 21.1% by 2035 and paid a median $135,880 in May 2025. For more careers like these, see the 25 most AI-proof jobs.

Will AI replace doctors and nurses?

Nurses are even further from replacement than doctors. Registered nurses are the largest health occupation in the BLS data, with about 3.47 million jobs in 2025, and they are projected to grow 5.6% by 2035, with roughly 180,800 openings a year. Nurse practitioners are projected to grow 41.0%, the fastest rate of any occupation in the data. Microsoft ranks registered nurses 460th of 785 for AI applicability, below family doctors, because so much of nursing is physical care at the bedside: turning a patient, starting an IV, noticing that someone looks worse than their chart says. AI will help nurses with charting and shift handovers, but it can’t do the rounds.

Will AI replace doctors in 10 years?

The official ten-year outlook says no. The BLS projections run from 2025 to 2035 and take automation into account, and every physician specialty is projected to grow. Searches like “will AI replace doctors in the future” or “in 50 years” go beyond what anyone can forecast credibly. What the evidence shows is which tasks go first, and in medicine that is paperwork, then pattern-spotting, with hands-on care and accountable judgment last.

What will change within ten years is how doctors train. Residents have always learned by writing notes, reading films and working through differential diagnoses, which are exactly the tasks AI now drafts. Medical students and residents should expect to be judged less on how fast they produce documentation and more on clinical judgment, communication and how well they check AI output. The early-career pressure we see in other fields, described in What jobs will AI replace?, shows up in medicine mainly in support roles rather than in physician jobs, which are protected by licensing and a long training pipeline.

How to AI-proof a career in medicine

The rule is the same for doctors, physician assistants and nurses: let AI handle the production, and own the judgment and the relationship. Here is how each type of task holds up:

For medical careers, five moves matter most:

1. Use AI for documentation, and check every note. A scribe that saves an hour a day is a real gain, but the signature on the note is yours, and so is any error in it.
2. Learn where AI tools fail. Know which tools your hospital uses, what they were trained on and where they miss. Clinicians who can judge AI output become more valuable, not less.
3. Invest in the human core. Physical examination, procedures, breaking bad news and shared decisions are the least exposed parts of medicine and the parts patients value most.
4. In image-heavy specialties, move toward the decisions. Radiologists and pathologists who work closely with treating teams, do procedures or lead the adoption of AI tools are in the strongest position.
5. In support roles, follow the information. Medical secretaries, coders and records staff can move toward health informatics, quality and compliance, where someone has to oversee the systems.

The same pattern of AI taking the drafting and people keeping the responsibility plays out in law, as we show in Will AI replace lawyers?. For a full 90-day plan to build the skills that hold their value, see our guide to the skills AI can’t replace.