For how we live.
Every article we have published on Artificial Intelligence, newest first.
In preregistered trials of more than 90,000 patients, telling people an algorithm flagged them as high-risk lifted vaccination by a few points — and saying the algorithm was involved changed nothing.
Computer-aided detection lifts polyp pickup in ordinary screening. In 757 patients having surveillance for an inherited cancer syndrome, a randomised trial could not show the same gain.
A final order published 17 September denies a petition that sought to exempt four classes of computer-aided detection, diagnosis and triage devices from 510(k) clearance. Makers must keep filing before they market.
A January 2025 draft guidance lays out the documentation the agency expects across an AI device's whole life, including bias checks across demographic groups and postmarket monitoring. It is not yet binding.
A deep-learning model reads the pancreas on non-contrast CT taken for other reasons. It reached very high accuracy in large tests — none of them a prospective screening trial.
A federal rule now requires certified electronic health records to publish a standard set of facts about each predictive algorithm they ship, including how it was validated and whether it was tested for fairness.
Software that interprets cardiotocography in labour lifts clinicians' accuracy in reader studies. The largest randomised trial of computerised interpretation, in 47,062 women, found no benefit for mothers or babies.
Companion chatbots eased loneliness as much as talking to a person in one set of studies — and the same research documents emotional dependence and users who are lonelier to begin with.
Under Article 6, AI that is or sits inside a device already needing an independent safety check counts as high-risk. A 70-study review finds the trouble is overlap with existing device law.
The Hypotension Prediction Index warns anaesthetists before blood pressure falls. Three 2026 trials find a plain mean-pressure alarm does the same job, and the algorithm's benefit tracks how hard clinicians treat.
TRICORDER put an AI-enabled stethoscope into UK primary care. Over a year it did not increase heart-failure diagnoses population-wide, even though the device, when actually used, flagged more disease.
Every low-dose CT for lung cancer also images the heart. A 323-scan study found an AI tool matched expert calcium scoring almost exactly — and separate data show why that number carries weight.
A randomised trial put a number on how much misinformation lowers vaccination intent. A systematic review mapped how common it is. And inoculation studies show the skill of spotting it can be trained at scale.
A multicentre reader study found AI raised early-career dentists' caries sensitivity from 67% to 82% without more false positives. Independent testing shows the same tools still miss early lesions.
Wrist and body sensors detect roughly 9 in 10 tonic-clonic seizures. A 192-patient validation against video-EEG found the harder focal seizures still slip past, or bury the alarm in false alerts.
Two 2026 studies posted high accuracy for AI reading Pap-smear cells and colposcopy images in settings short of pathologists. Both were validated internally, which is where such numbers usually shrink.
A final order files AI tools that flag possible cardiovascular disease as class II with special controls, letting similar triage software clear via 510(k). The controls require real-world, subgroup-tested validation.
An autonomous AI for diabetic retinopathy hit 87.2% sensitivity in primary care. A smartphone acuity test agreed with a reference chart to within 0.07 logMAR. Both are triage, not diagnosis.
Commercial software reads an X-ray for breaks about as well as a radiologist. A prospective paediatric emergency study found that accuracy did not translate into fewer missed-fracture recalls.
A year of alerts at one hospital shows the safety pop-ups clinicians dismiss almost automatically — and why the field still cannot agree on how to measure the problem.
In a prospective African validation, TB-detection AI matched radiologists but neither hit the WHO sensitivity target. In an Indian field programme, AI-led screening lifted case-finding sharply.
A predetermined change control plan is the FDA's answer to a model that keeps learning: pre-approve a bounded set of future changes, and the manufacturer can ship them without returning for a fresh review.
Systems that give a diagnosis with no clinician in the loop post high sensitivity in trials. In real-world use, one in six images was too poor to grade — and the strongest numbers come from company-run studies.
Pooled diagnostic accuracy is high in specialist settings, but performance falls on Fitzpatrick IV–VI skin and on smartphone images, and most studies never report skin tone at all.
A deep-learning system drew tumour targets and organs at risk automatically. It saved time and improved consistency — but oncologists still reviewed and edited the contours before treatment.
The first randomised trial to report interval cancers found AI non-inferior to double reading, with higher sensitivity at equal specificity. Across three European programmes the extra detection is about one per 1000.
One found doctors stopped using it as shifts got busier. The other, in 9,691 Kenyan patients, found it safe but no better than the electronic record alone. Neither supports deployment.
A Polish study found endoscopists' unaided polyp-detection fell after AI arrived. A Norwegian trial that tracked doctors before, during and after AI use found no such deskilling.
Algorithms that read short voice recordings can pick up acoustic signs of some conditions. In the largest and most careful studies the signals are real but narrow, sex-specific, and often fail to generalise.
Researchers built an AI pipeline to convert decades of narrative reports into one modern TNM staging system, reaching 95% accuracy for tumour classification in an expert-labelled test set.
A Danish stepped-wedge trial gave endoscopists automated feedback on their technique after every procedure. Adenoma detection rose from 43.4% to 48.6% — a different tool from real-time polyp AI.
Trained on 1.7 million ECGs paired with clinicians' report text, ECG-CLIP reached the same accuracy as the best comparator with about 90% less training data — a bid at the field's labelling bottleneck.
In 6,772 users of a US health system's AI triage tool, people engaged about twice as often when the AI matched what they already planned — raising a hard question about what the tools are steering.
A review of 41 studies found that the great majority of AI medication-adherence prediction models carried high risk of bias, and that fancier algorithms did not reliably predict better.
Across 24 official German licensing exams, the best model answered 99.31% of first-exam items correctly. On items containing an image, the error rate rose several-fold, against 1.24x for students.
A survey of seven countries found the youth mental health app market dominated by paid journaling and meditation apps. The AI-powered ones show early promise on symptoms — and none had checked for harm.
A researcher who expected the evidence base to be thin says even she was surprised by how thin. Most cleared devices never appear in a registered clinical trial at all.
A new discussion paper proposes a two-axis risk framework and a physician-training analogy for evaluating generative AI devices. It is not a rule, and the agency is asking what one should look like.
Endometriosis takes up to a decade to diagnose. A modelling study estimated a digital symptom checker could cut that by 4.36 years — but the tools still lack the external validation to trust in practice.
Where AI clot-detection was tested head-to-head, it still missed more emboli than radiology trainees, especially small peripheral ones. Its clearest win was cutting time to diagnosis.
A Nature Reviews Drug Discovery audit says the problem isn't the models. They were built to be validated rather than used, and benchmarked against the wrong thing.
Most wellness apps make no disease claim and face no federal review. Into that gap, US states are writing their own AI-in-mental-health laws — four already diverge, leaving users unevenly protected.
An umbrella review of 42 reviews found only two wearable measures — heart rate and atrial fibrillation — with comparably strong support. Getting any device into a trial is harder than buying it.
Two 2026 studies found people with no ultrasound background produced diagnostic-quality echocardiograms after about a day of AI-guided training. Quality varied sharply by view, finding, and who was holding the probe.
42 orthopedic physicians diagnosed 40 rare diseases twice, once alone and once after seeing AI suggestions. Accuracy jumped 20 to 26 points, but the same-day design leaves memory unaccounted for.
A Lancet Digital Health scoping review found the field's fairness metrics fragmented and rarely clinically validated. A second review found that most studies don't measure fairness at all.
A review of Paige Prostate Detect and Ibex Prostate Detect finds the tools mainly help less-specialized pathologists, and warns performance shifts when a tool meets a new population.
Two 2026 reviews make the case that human-derived models can replace animal testing. A third paper argues the field is checking the biology and the algorithm separately, and calling that assurance.
Two 2026 meta-analyses pooled deep-learning systems reading retinal images for AMD. The accuracy looks near-perfect, but almost all of it comes from algorithms tested on their own data.
A paper in the journal Resuscitation lays out how a large language model could help dispatchers recognize cardiac arrest and coach CPR in real time — while acknowledging the concept still needs to prove it saves lives.
Pooled across 48 trials, chatbots cut depression, anxiety and stress by small amounts. But a review of purpose-built mental health chatbots found crisis-referral protocols underdeveloped and missed suicidal ideation.
A small proof-of-concept study pitted several AI systems against gastroenterologists and emergency physicians on cholangitis exam questions. The gap between the best and worst AI performers was enormous.
The agency granted its first authorization for a software-aided adjunctive diagnostic device in wound assessment, a category built for tools that analyze a wound optically.
A study running 5.3 million evaluations through nine large language models found eligibility judgments were largely stable across identity labels — except when a patient vignette mentioned homelessness.
DxDirector-7B beat human physicians on a benchmark of complex diagnostic cases while requesting far fewer tests. Its own authors say it isn't ready for high-risk or emergency cases.
A secondary analysis of the VITAL-AF trial reports a screening benefit only in the top risk decile — with a confidence interval whose lower bound sits at 0.01.
GE's Critical Care Suite gains an algorithm that flags misplaced enteric tubes on a chest X-ray — a complication that reviews of the practice have linked to respiratory harm and, in some cases, death.
The clearance is the third in a growing family of machine-learning notification algorithms built to spot serious cardiac conditions from a routine 12-lead ECG, a test most patients already get.
A new audit of China's regulatory record finds a market concentrated in radiology, dominated by deep learning, and clustered in four cities. It also finds the approval curve flattening.
A 1,298-person randomized study found people using chatbots to work through medical scenarios did no better than people without them — even though the same chatbots, tested alone, got the right answer most of the time.
Researchers ran 960 responses through ChatGPT Health using clinician-written vignettes. Failures clustered at both extremes, and crisis safeguards activated unpredictably.
A February theme issue documents faster notes, happier clinicians and enterprise rollouts to thousands. It also contains an editorial asking the question none of the studies answer.
Ten studies, wide confidence intervals, and factual error rates of 26 to 36 percent in AI-drafted documentation. The review's own conclusion is that the evidence is preliminary and highly uncertain.
AlphaGenome takes a million base pairs of DNA and predicts what the sequence does. In Nature on January 28, it matched or beat the best existing models in 25 of 26 variant-effect evaluations.
Clinicians rated empathy at 4.6 out of 5 and quality of information at 2.7. Every chatbot produced at least one piece of guidance judged inappropriate, overstated or inaccurate.
The landmark audit found the right diagnosis listed first in 34% of cases and correct triage in 57%. A newer heart-attack test exposes the dangerous edge: most atypical cases missed, and women.
A transformer model trained on 17,119 scans beat expert and non-expert examiners on every metric and cut simulated referrals by 63%. The catch is who built it and the validation the rest of the field still lacks.
The EAGLE trial ran colonoscopy AI off-site over a network and aimed it at the lesions that matter. It reports a threefold gain in serrated lesions — in a field whose main US guideline recommends nothing.
One agency asked how to measure AI devices after they are deployed. Another asked how to speed adoption up. The offices that would answer the first question are losing staff.
The headline finding is noninferiority. The more useful finding is what the shared denominator was — and how wide a gap the trial was designed to tolerate.
England ran the closest thing to a national test of clinical AI anyone has published. The software sites gained more, but the comparison sites were improving on their own, and that gap is the whole result.
The WISeR model starts on 1 January in six states. The contractors running it are paid out of the savings their determinations produce — the design feature doctors keep pointing at.
Researchers showed open-source protein design software could produce variants of proteins of concern that biosecurity screening tools missed, then wrote and deployed patches before publishing.
A census of 691 AI-enabled devices cleared through 2023 found demographic data missing almost everywhere, three summaries reporting patient outcomes, and 113 recalls driven mostly by software.
EyeFM was tested as an assistant to 16 ophthalmologists screening 668 high-risk patients in China. Patients in the AI arm also followed referral advice more often.
Google's PH-LLM beat sampled human experts on multiple-choice tests but only matched them on real cases. A new reporting checklist published the same month explains why such claims are hard to compare.
EchoNext scored 77.3% accuracy on a 150-ECG set where 13 cardiologists averaged 64.0%. Its authors released the model weights and a 100,000-ECG labelled dataset alongside the paper.
A draft-reporting model cut documentation time 15.5% across 23,960 radiographs without changing report quality. A separate reader study found AI assistance raised prostate MRI accuracy by 3.3 percentage points.
Rentosertib was safe over 12 weeks in 71 patients with idiopathic pulmonary fibrosis. A lung-function signal appeared at the highest dose, in a trial not designed to prove efficacy.