This Viewpoint discusses speed and safety in pediatric artificial intelligence.

This quality improvement study assesses the vulnerability of leading commercial large language models to invisible text injection manipulation in simulated medical peer review.

JAMA+ AI Associate Editor Yulin Hswen, ScD, MPH, spoke with Cecilia Lee, MD, MS, a professor of ophthalmology at Washington University in St Louis, for JAMA+ AI Conversations.

This quality improvement study assesses the association of a natural language processing tool classifying patient portal messages as high acuity with clinician time to first read in a large, integrated health care system.

This Special Communication explores the use of artificial intelligence in mental health care delivery, outlining the potential benefits and negative consequences, and suggests strategies that may help mitigate the risks.

This cohort study evaluates use of a machine learning algorithm with molecular imaging to analyze imaging data during lung cancer surgery to determine the malignant potential of nodules.

This Viewpoint describes potential risks of using generative AI–driven ambient scribes to automate more intensive billing.

This cohort study evaluates physician revenue, patient volumes, and claim denials among adopters and nonadopters of artificial intelligence–based clinical documentation tools in one health system.

This Medical News article discusses the growing use of generative artificial intelligence chatbots as sources of mental health support.

This qualitative study evaluates ambient scribe technology using standardized patient encounters in a simulated clinical environment, combining a validated measure of documentation quality with qualitative insights.

This cohort study evaluates whether artificial intelligence–driven length-of-stay forecasting can facilitate and help stabilize postoperative hospital bed management and capacity.

JAMA+ AI Associate Editor Yulin Hswen, ScD, MPH, spoke with Jonathan Chen, MD, PhD, and Ethan Goh, MD, MS, of Stanford University in Stanford, California, for JAMA+ AI Conversations.

This prognostic study investigates whether artificial intelligence (AI) prediction models for in-hospital mortality using International Classification of Diseases diagnostic codes are associated with inflated performance metrics via label leakage.

This diagnostic study examines the accuracy of machine learning with deep transfer learning algorithms to predict post–cochlear implant spoken language development in children with bilateral sensorineural hearing loss.

This prognostic study develops and validates a machine-learning model that predicts rupture risk of unruptured intracranial aneurysms.

Nature Machine Intelligence, Published online: 14 January 2026; doi:10.1038/s42256-025-01152-1

Large language models are starting to be used in safety-critical tasks such as controlling robots. Zhou et al. present LabSafety Bench, a benchmark evaluating the ability of large language models to identify hazards and assess laboratory risks.

Nature Machine Intelligence, Published online: 31 December 2025; doi:10.1038/s42256-025-01167-8

Deep generative models that learn intermediate surface-current maps, rather than layouts directly, offer a more stable route to inverse design of tunable and stacked metasurfaces.

Nature Machine Intelligence, Published online: 31 December 2025; doi:10.1038/s42256-025-01166-9

The reusability of AlphaTensor-Quantum is tested and the method is extended to optimize a broad range of quantum circuits without retraining, achieving greater T-count reductions and demonstrating generalizable and efficient quantum circuit optimization.

Nature Machine Intelligence, Published online: 31 December 2025; doi:10.1038/s42256-025-01170-z

He et al. present a parameter-efficient fine-tuning method for single-cell language models that improves performance on unseen diseases, treatments and cell types.

Nature Machine Intelligence, Published online: 31 December 2025; doi:10.1038/s42256-025-01161-0

Xiao et al. present GITIII, a lightweight and interpretable graph transformer for inferring spatial single-cell-level interactions and quantifying the influence of neighbouring cells on the gene expression of receiver cells in spatial transcriptomics.

Nature Machine Intelligence, Published online: 31 December 2025; doi:10.1038/s42256-025-01162-z

Metasurface design driven by AI faces challenges, such as extrapolation to unexplored performance regimes. MetaAI, a physics-aware current-diffusion framework, is introduced to advance metamaterial discovery from interpolation to extrapolation.

Nature Machine Intelligence, Published online: 31 December 2025; doi:10.1038/s42256-025-01160-1

Morehead et al. introduce the benchmark PoseBench and evaluate the strengths and limitations of current AI-based protein–ligand docking and structure prediction methods.

Nature Machine Intelligence, Published online: 31 December 2025; doi:10.1038/s42256-025-01163-y

A multimodal deep learning model combines molecular sequence, structure and biochemical properties to predict immunogenicity in an interpretable way, providing a framework for smarter molecular prediction and hypothesis generation.

npj Digital Medicine, Published online: 23 January 2026; doi:10.1038/s41746-026-02358-2

Prospective real-world implementation of deep learning systems in healthcare: a systematic review guided by implementation science

npj Digital Medicine, Published online: 23 January 2026; doi:10.1038/s41746-025-02309-3

Uncertainty modeling in multimodal speech analysis across the psychosis spectrum

npj Digital Medicine, Published online: 23 January 2026; doi:10.1038/s41746-026-02361-7

HMC-transducer: hierarchical mamba-CNN transducer for robust liver tumor segmentation

npj Digital Medicine, Published online: 22 January 2026; doi:10.1038/s41746-025-02321-7

A systematic review of AI for predicting glaucoma progression: challenges and recommendations towards clinical implementation

npj Digital Medicine, Published online: 22 January 2026; doi:10.1038/s41746-026-02372-4

Early diagnosis of axial spondyloarthritis in primary care using multi-agent systems

npj Digital Medicine, Published online: 22 January 2026; doi:10.1038/s41746-026-02368-0

Personalized supervised and unsupervised intracranial sleep decoding during deep brain stimulation

npj Digital Medicine, Published online: 22 January 2026; doi:10.1038/s41746-026-02363-5

Large language models improve transferability of electronic health record-based predictions across countries and coding systems

npj Digital Medicine, Published online: 22 January 2026; doi:10.1038/s41746-026-02380-4

Evaluation of large language models for diagnostic impression generation from brain MRI report findings: a multicenter benchmark and reader study

This study evaluated whether continual learning could improve the generalization of an artificial intelligence model for assessing endotracheal tube placement on chest radiographs across diverse hospitals worldwide. Using data from 23 hospitals in 12 countries, the researchers found that models trained through continual learning consistently outperformed baseline and single-hospital fine-tuned models, demonstrating stronger and more reliable performance across varied clinical environments.

This study evaluated the ability of generative pretrained transformer 4o to grade short-answer responses from first-year medical students compared with human graders. Using iterative, pedagogy-guided prompt refinement, GPT-4o achieved moderate agreement with human scores, suggesting that artificial intelligence–assisted grading could help reduce the burden of assessing open-ended questions in medical education.

This editorial explores the utility of multiple-choice questions and short-answer questions for assessment in medical education. It also addresses the extent to which the study by Kuling et al. demonstrates whether a large language model can adequately grade short-answer questions.

This perspective article explores what constitutes a high-value application in medical artificial intelligence.

Like all science, regulatory science can and should evolve and improve. That is the encouraging message conveyed in the article by Scheufen Tieghi and colleagues in this issue of NEJM AI

In this issue, Tieghi and colleagues outline a federal road map to accelerate the shift from animal testing to new approach methodologies; however, the authors argue that the operationalization of NAMs is currently hindered by a lack of benchmarking infrastructure.

This Policy Corner reviews the rapidly accelerating shift in United States regulatory policy and scientific practice toward human-centered alternatives to animal testing in drug development. It outlines the coordinated initiatives of the U.S. Food and Drug Administration and National Institutes of Health to validate and implement new approach methodologies — including artificial intelligence–driven modeling, organoids, and organ-on-chip systems — to create a more ethical, efficient, and human-relevant framework for preclinical safety and efficacy testing.

This paper introduces VeriFact, an artificial intelligence system designed to verify the factual accuracy of AI-generated clinical documents using information from patients’ electronic health records. By combining retrieval-augmented generation with large language models acting as evaluators, the study demonstrates a scalable approach to ensuring reliability and trustworthiness in automated medical documentation.

This feasibility case study evaluated the first prospective bedside deployment of the Discharge after Surgery Using Artificial Intelligence artificial intelligence model for predicting safe surgical discharge. The study demonstrated that, although DESIRE achieved strong predictive performance and operational feasibility, its reliability was constrained by data quality and governance challenges — highlighting that operational readiness, not model accuracy alone, determines clinical feasibility and safety.

This perspective challenges the widespread use of disease classification in medical artificial intelligence, arguing that reducing continuous disease complexity to discrete classes fundamentally misrepresents pathophysiology and limits precision medicine. The authors advocate for continuous disease assessment approaches that model fine-grained disease characteristics, patient trajectories, and treatment responses rather than forcing complex conditions into artificial diagnostic buckets.

This article highlights the emerging public health and regulatory dilemmas of “relational” artificial intelligence, warning that emotionally dependent attachments to AI companions may potentially carry significant risks that are not yet well understood. We argue for proactive clinical, research, and policy responses to anticipate harm and maximize societal benefits before loosely regulated market forces potentially dictate the trajectory of this technology.

This perspective situates contemporary concerns about artificial intelligence–driven cognitive offloading within a long history of debates over how reference tools — from early medical handbooks to digital systems — reshape perspectives on clinical expertise. By tracing how definitions of skill have evolved alongside these technologies, it reveals that shifts in medical cognition reflect changing historical contexts rather than purely technological causes.

Generative AI is making it astonishingly easy to create scientific fakery that looks real: convincing data, tidy plots, even entire studies that slip past automated checks and human reviewers. In this editorial, the author describes a deliberately fabricated dataset and analysis, built with help from an AI model that fooled standard anomaly detectors. The episode highlights a growing problem: technical fixes like blockchain may offer reassurance, but they can’t solve the deeper cultural pressures in science that reward speed and novelty over care and verification. What might help is the harder work of valuing replication, transparency, and rigor, as AI can amplify whatever incentives we set.

This clinical case study evaluated the real-world implementation of an artificial intelligence triage system for detecting for detecting features of lung cancer lung cancer on chest x-rays across a multihospital National Health Service trust in South West London. Following deployment, the AI-assisted “radiographer-led same-day computed tomography pathway” reduced the median time from chest x-ray to CT report from 6.0 to 3.6 days without increasing overall CT volume, demonstrating the potential of AI triage to accelerate diagnosis in urgent cancer pathways.

This letter reflects on the editorial “Humanity’s Next Medical Exam” and highlights the need for AI systems that restore physician presence and align technology with human values. It argues that doctors who code must focus on the 60% of clinical work lost to inefficiency and administrative burden.

For decades, the pharmaceutical industry has operated on a singular, rigid principle: the “lock-and-key” model. If a protein is the lock driving a disease like Alzheimer’s or cancer, scientists must find a chemical “key” that fits perfectly into its static, three-dimensional structure to turn it off. The problem? Nearly one-third of the human proteome doesn’t ...;

Once a pharmaceutical product enters the market, its legal exposure is far from over. In many cases, the most significant safety insights emerge only after widespread, real-world use. This is where post-market drug monitoring becomes central—not only to public health oversight, but to legal accountability. As reporting systems, data collection tools, and monitoring technologies grow ...;

Late one evening in Oslo, a general practitioner finishes seeing her last patient, but her work is far from over. Like many doctors, she faces another hour or two of typing up patient notes and finishing documentation. She’s not alone – 69% of physicians report spending too much time on after-hours charting, and 62% say ...;

Artificial intelligence is rapidly transforming how people search for medical information, choose doctors, and evaluate healthcare providers online. Traditional search engine optimization is no longer enough for medical brands that want to remain visible, trusted, and competitive. Today, the real opportunity lies in Medical AI SEO, a specialized approach that helps healthcare providers appear accurately ...;

Much like training an advanced AI system, intense workouts push your body beyond comfort zones so it can adapt, grow stronger, and perform better over time. However, continuous high-intensity training can lead to muscle soreness, reduced energy, and mental exhaustion. Fatigue is natural, but managing it intelligently helps you stay consistent, motivated, and injury-free. This ...;

Healthcare has entered a phase where data is no longer a byproduct of care—it is a driving force behind how care is delivered, measured, and improved. Medical devices, once designed primarily to perform a single mechanical or clinical function, are now evolving into intelligent systems capable of generating, transmitting, and responding to real-time data. This ...;

Medical imaging has always been central to diagnosis, but the growing volume and complexity of visual data are pushing traditional analysis methods to their limits. Clinicians today must interpret thousands of images generated by advanced scanners, often under time pressure and with limited resources. This challenge has created a gap between data availability and actionable ...;

Heart disease remains the silent assassin of the American healthcare system, claiming one in three lives and costing the economy hundreds of billions annually. Yet, for decades, the tools cardiologists use to catch it have been stuck in the pager era. Patients sent home for monitoring are often strapped with wired and bulky monitors that ...;

Artificial intelligence is reshaping healthcare by improving mobility aids. These advancements make aids more user-friendly and efficient. AI continues to offer new opportunities for enhancing mobility solutions. The integration of artificial intelligence into healthcare has led to significant advancements, particularly in the design and functionality of medical mobility scooters and other mobility aids. By leveraging ...;

We’re launching a new 27 billion parameter foundation model for single-cell analysis built on the Gemma family of open models.

Google announces its first cohort of 25 startups using AI to improve public services.

Learn more about Google’s partnership with NBA player Stephen Curry.

Doctors, developers and researchers gathered in Paris to prototype new medical solutions using Google’s AI models.

This field guide and investment support AI’s potential in evidence-based mental health interventions and research.

Here are Google’s latest AI updates from June 2025

Google Health and Google Cloud are collaborating with Taiwan’s National Health Insurance Administration.

Read Ruth Porat's remarks on AI and cancer research at the American Society of Clinical Oncology.

Health systems and hospitals spent more than $1 billion on the technology in 2025, with ambient AI and coding/billing programs...

The health system has posted impressive patient safety improvements, including a 67% reduction in serious safety events from June 2019...

WVU Medicine's revenue cycle leader discusses how AI helps prevent denials, and where basic automation is still the smarter choice.

At Baptist Health, AI governance demands a deep collaboration across IT and clinical teams, strict contract metrics, and essential exit...

Allina Health has determined that use of AI scribes reduces clinician burnout and helps clinicians maintain their full-time equivalent performance,...

Barry Stein, Hartford HealthCare?s Chief Clinical Innovation and Medical Informatics Officer and founder of the Center for AI Innovation in...

Here are five urgent questions that will determine revenue cycle strategy in 2026, from the rise of autonomous AI to...

The state is the first to launch a pilot that enables patients living with chronic health concerns to refill their...

The researchers found that clinician use of an AI scribe reduced burnout, decreased the amount of time spent generating clinical...

Aided by AI and better EHRs, healthcare leaders are integrating clinical trial recruitment into care management and coordination.

Allina Health has determined that use of AI scribes reduces clinician burnout and helps clinicians maintain their full-time equivalent performance,...