Breakthrough tracker
17 AI-in-medicine breakthroughs, each rated for how far away it really is. Search, or filter by evidence strength or area.
Results
- Cancer
AI help for reading mammograms
In a large Swedish trial (MASAI), mammogram screening supported by AI found more cancers than standard reading by two radiologists, and cut radiologists' reading workload by about 44%. It's strong trial evidence, but this way of screening isn't yet routine in most screening programmes.
Regulator record: US, EU, UK · Updated · Track: US 4/5, EU 4/5, UK 3/5, AU 3/5 - Cancer
AI that predicts five-year breast cancer risk from a mammogram
In May 2025 the US FDA granted Clairity's De Novo request for Allix5: software that estimates a woman's chance of breast cancer in the next five years from a routine screening mammogram. It's authorised in the US only. It gives a prediction, not a diagnosis.
Regulator record: US · Updated · Track: US 4/5, EU 3/5, UK 3/5, AU 3/5 - Lungs and chest
AI that double-checks chest X-rays
Annalise CXR is AI software that flags possible findings on chest X-rays for radiologists to check. A version of it has FDA 510(k) clearance in the US for triage of some findings, and a large Australian radiology network reported using it in 2021. A radiologist still reads and reports every scan.
Regulator record: US, EU, UK · Updated · Track: US 4/5, EU 4/5, UK 3/5, AU 3/5 - Bowel and gut
AI during colonoscopy, and the 'deskilling' question
AI that highlights possible polyps during colonoscopy has been authorised in the US since 2021. But a 2025 Polish study found doctors who got used to AI spotted fewer precancerous growths when working without it. AI help is real; so is the risk of leaning on it.
Regulator record: US, EU · Updated · Track: US 4/5, EU 4/5, UK 3/5, AU 3/5 - Lungs and chest
An AI-designed drug for lung scarring (rentosertib)
Rentosertib, a drug whose target and molecule were found using generative AI by Insilico Medicine, was tested in 71 people with idiopathic pulmonary fibrosis in a 12-week phase 2a trial. It looked safe enough to keep testing, though some people stopped because of liver problems or diarrhoea. Lung function (FVC) improved at the top dose, but that was an exploratory measure in a small, short, company-run safety trial in China only. It isn't approved in the US or EU, and we haven't yet searched the UK and Australian registers. A larger phase 3 trial started in 2026; results, and any approval, are likely years away.
No regulator approval recorded · Updated · Track: US 2/5, EU 2/5, UK 2/5, AU 2/5 - Drug discovery
AlphaFold 3: predicting how the molecules of life fit together
AlphaFold 3, published in Nature in May 2024 by Google DeepMind and Isomorphic Labs, predicts how proteins interact with DNA, RNA and drug-like molecules. Demis Hassabis and John Jumper of Google DeepMind shared half the 2024 Nobel Prize in Chemistry for AlphaFold2. It's widely used in research, but we're not aware of any medicine from it reaching patients yet.
No regulator approval recorded · Updated · Track: US 1/5, EU 1/5, UK 1/5, AU 1/5 - Prediction and risk
Delphi-2M: AI that forecasts risk of 1,000+ diseases
Delphi-2M, a GPT-style model published in Nature in September 2025, learned from health records of about 400,000 UK Biobank participants to predict rates of more than 1,000 diseases. It held up on 1.9 million Danish records. It's early research and not ready for clinical use.
No regulator approval recorded · Updated · Track: US 1/5, EU 1/5, UK 1/5, AU 1/5 - Diagnosis (general)
AMIE: Google's diagnostic chatbot research
Google's AMIE is a research AI that takes a medical history by text chat. In a study published in Nature in 2025, it was rated as more accurate than primary-care doctors on 159 simulated cases with actors as patients. That study used no real patients; a later real-patient study is tracked as TR-016.
No regulator approval recorded · Updated · Track: US 1/5, EU 1/5, UK 1/5, AU 1/5 - Diagnosis (general)
Microsoft's MAI-DxO: AI 'panel of doctors' on hard cases
In June 2025 Microsoft reported MAI-DxO, an AI system that works through difficult published case studies step by step. It scored much higher than doctors in a test Microsoft designed. It's a preprint, not peer-reviewed, and was tested on case records, not patients.
No regulator approval recorded · Updated · Track: US 1/5, EU 1/5, UK 1/5, AU 1/5 - Infections
Generative AI designs new antibiotics against superbugs
MIT researchers used generative AI to design brand-new antibiotic molecules, published in Cell in August 2025. Two candidates worked against drug-resistant gonorrhoea and MRSA in the lab and in mice. It's early research: no human has taken them.
No regulator approval recorded · Updated · Track: US 1/5, EU 1/5, UK 1/5, AU 1/5 - Skin
DERM: AI skin-lesion triage that can send some likely-harmless moles off the urgent cancer pathway
DERM is AI that assesses photos of skin lesions referred on the urgent skin cancer pathway. In May 2025 NICE conditionally recommended it for NHS use in England for three years while more evidence is gathered. Evidence is less certain for black and brown skin, so NICE says those patients get an extra review by a healthcare professional. It doesn't replace checking your own skin.
Regulator record: UK · Updated · Track: US 3/5, EU 3/5, UK 3/5, AU 3/5 - Cancer
ArteraAI Prostate: AI that reads biopsy slides to forecast prostate cancer outcomes
On 31 July 2025 the US FDA granted Artera's De Novo request for ArteraAI Prostate, software that analyses digitised biopsy slides to estimate long-term outcomes for men with non-metastatic prostate cancer. It's authorised in the US only and supports, rather than replaces, treatment decisions.
Regulator record: US, UK · Updated · Track: US 4/5, EU 3/5, UK 3/5, AU 3/5 - Heart
AI stethoscopes in GP surgeries: what a 205-practice NHS trial found
The TRICORDER trial gave AI-enabled stethoscopes to NHS GP practices to help spot heart failure, atrial fibrillation and valve disease. Its main result was neutral: practices with the device didn't diagnose more heart failure overall. Many practices used it little. Where it was used, more diagnoses were recorded, but that comparison can't show the device caused them.
Regulator record: US · Updated · Track: US 4/5, EU 3/5, UK 3/5, AU 3/5 - Diagnosis (general)
AI scribes that write up your appointment: the first randomised trials
In November 2025 NEJM AI published two randomised trials of 'ambient' AI scribes, which listen to a consultation and draft the doctor's notes. In one trial, one of the two scribes modestly reduced time spent writing notes; in the other, AI reduced clinician exhaustion scores. Doctors reported occasional clinically significant inaccuracies, so notes still need checking.
No regulator approval recorded · Updated · Track: US 2/5, EU 2/5, UK see page, AU see page - Genetics
AlphaGenome: AI that predicts how DNA changes might affect gene activity (research tool)
In January 2026 Google DeepMind published AlphaGenome in Nature: a model that reads up to a million letters of DNA and predicts how small changes (variants) affect how genes are switched on and off. It's a research tool for scientists and isn't used to diagnose patients.
No regulator approval recorded · Updated · Track: US 1/5, EU 1/5, UK 1/5, AU 1/5 - Diagnosis (general)
AMIE: Google's research chatbot tested with real patients before GP visits
In a Lancet study published 8 October 2026, 100 adults at one Boston primary-care clinic text-chatted with Google's research AI, AMIE, before an urgent appointment, and 98 then had the visit. A doctor watched every chat live, and none had to be stopped under the safety rules. It was a small, Google-funded feasibility study with no comparison group.
No regulator approval recorded · Updated · Track: US 2/5, EU 2/5, UK 2/5, AU 2/5 - Eyes
AI eye screening in Tanzania: more people reached the eye clinic
In a randomised trial of 2,378 people with diabetes at 24 Tanzanian hospitals, AI eye screening with an on-the-spot result and face-to-face counselling roughly doubled clinic attendance among the 904 who needed referral (35.9% vs 18.2% within 8 weeks). It measured attendance, not treatment or sight saved.
No regulator approval recorded · Updated · Track: US 2/5, EU 2/5, UK 2/5, AU 2/5
No breakthroughs match. Try a different word or clear the filters.
Policy watch
Government and regulator decisions that shape how AI reaches patients. Not rated, and not counted in the total above.