LLMs are artificial-intelligence systems trained on enormous collections of text and other data to predict and generate sequences of language. In a clinical environment, that ability can be used to summarize patient records, draft medical notes, explain complex information, retrieve relevant research and assist with diagnostic reasoning. Some models are also being connected to imaging systems, laboratory data and electronic health records. This integration gives them access to highly sensitive information and places their outputs closer to real medical decisions. The review notes that many healthcare workers are already using such tools informally, sometimes without institutional approval, technical safeguards or clear instructions about responsibility. This unofficial “shadow use” makes it difficult for hospitals to know which systems are being used, what information is being entered and whether the resulting recommendations influence patient care.

The authors emphasize that risk is not confined to the moment when a clinician receives an AI-generated answer. It can emerge throughout the entire lifecycle of a system, beginning with model design and continuing through data collection, training, deployment, software updates and everyday use. A model may inherit weaknesses from its training data, including incomplete representation of populations, outdated medical knowledge or systematic biases. Data used to train or adapt a model can also be deliberately manipulated in an attack known as data poisoning. By inserting carefully crafted examples into a training set, an attacker may influence how the model behaves later. Because modern LLMs are complex and difficult to interpret, identifying the precise cause of a dangerous response can be challenging even after an incident has occurred.

Another cybersecurity threat involves prompt injection. In this type of attack, hidden or misleading instructions are embedded in text supplied to the model. The instructions can redirect the model away from its intended task, cause it to disclose information or make it ignore clinically relevant evidence. In a medical setting, a compromised prompt might cause an AI system to overlook an abnormality in a pathology report or fail to identify a tumor that is visible in a tissue sample. Prompt injection can be especially difficult to prevent when an LLM processes material from multiple sources, such as clinician instructions, patient records, web pages and laboratory reports. If the system treats all text as equally authoritative, malicious content may be mistaken for a legitimate command. Beyond the model itself, weaknesses in hospital networks, application programming interfaces and cloud infrastructure could expose patient data or interrupt essential clinical services.

The most familiar model-inherent hazard is the phenomenon commonly called hallucination: the generation of information that sounds convincing but is false, unsupported or invented. LLMs do not retrieve truth in the human sense. They generate statistically plausible sequences based on patterns learned during training, unless they are connected to carefully controlled databases or retrieval systems. Even retrieval-augmented systems can produce errors if the source material is incomplete, poorly indexed or misinterpreted. In medicine, a fabricated citation, incorrect drug interaction or inaccurate diagnostic explanation could have immediate consequences. The danger is amplified by the fluent and confident style of many models, which can make a wrong answer appear more reliable than a cautious human response.

The review also examines how human behavior can turn technical imperfections into clinical harm. Clinicians may develop automation bias, placing too much trust in an AI recommendation simply because it was produced by a sophisticated system. This can reduce the likelihood that an output will be independently checked, particularly in busy environments where staff face time pressure and heavy workloads. The opposite problem can occur when an AI system is used to confirm an existing belief. If a clinician suspects a particular diagnosis, a model may be prompted in a way that encourages it to support that assumption, reinforcing confirmation bias rather than challenging it. LLMs can also adapt their language to the apparent expectations of the user, producing answers that are more agreeable than accurate. Long or complicated conversations create further risks because errors can accumulate across multiple exchanges and become difficult to trace.

Privacy and accountability add another layer of complexity. Many commercially available AI services are hosted outside the healthcare institution, meaning that hospitals may have limited control over where patient information is processed, how long it is retained and whether it is used for further system development. Even when providers promise data protection, healthcare organizations must understand the technical and legal arrangements governing access, storage and deletion. De-identification can reduce privacy risks, but it is not a universal solution: detailed clinical narratives may contain enough unusual information to make individuals identifiable when combined with other data. At the same time, responsibility can become blurred when a recommendation is produced by a model, reviewed by a clinician and embedded in software supplied by a third party. The authors argue that clinical accountability must remain explicit rather than being dispersed across an opaque technical chain.

To reduce these dangers, the researchers call for security to be built into AI development from the beginning rather than added after deployment. This includes carefully curating training and fine-tuning data, testing models against adversarial attacks, documenting known limitations and evaluating performance on clinically representative cases. Hospitals should monitor systems continuously instead of treating approval as a one-time event. Monitoring can include tracking error patterns, unexpected changes in performance, suspicious user activity and the effects of software updates. The authors recommend clear institutional rules defining which tools may be used, what data may be entered, when human review is mandatory and who is responsible for responding to incidents. They also propose local teams dedicated to supervising AI in clinical practice, supported by centralized AI Security Operations Centers, or SOCs, that could identify threats across institutions and coordinate responses to attacks or failures.

The review further argues that regulation must evolve alongside rapidly changing AI technology. Only a small proportion of AI systems are formally approved as medical devices, while many general-purpose models can influence healthcare without fitting neatly into existing regulatory categories. Traditional frameworks were largely designed for products that remain stable after approval. LLMs, by contrast, may change through updated training data, modified instructions, new connected tools or silent software revisions. A system that performs well during an initial evaluation may behave differently months later or when used in a new hospital with a different patient population. The authors therefore support regulatory approaches based on continuous evaluation, post-deployment surveillance and transparent reporting of significant changes. Stephen Gilbert, Professor of Medical Device Regulatory Science at TU Dresden, says that oversight and technological development must be more closely integrated if patients and health systems are to benefit safely from these tools.

The researchers stress that safer medical AI will require cooperation among engineers, clinicians, cybersecurity specialists, regulators, ethicists and patients. Human oversight remains essential, but oversight must be meaningful: clinicians need sufficient training, enough time to review AI outputs and access to information about how systems were evaluated. Institutions should also make responsible use easier than unofficial use by providing approved tools, practical guidance and channels for reporting errors without fear of punishment. “AI is already being used in healthcare, often without formal oversight,” says Jakob N. Kather, Professor of Clinical AI at the Else Kröner Fresenius Center. “The key question is how to implement these systems in a way that is transparent, robust, and aligned with clinical responsibility.” The review’s central message is not that LLMs should be kept out of medicine, but that their deployment must be treated as an ongoing safety process. Powerful language models may become valuable clinical partners, but only if hospitals continuously test, monitor and govern them with the same seriousness applied to other technologies that can affect human life.

Subject of Research: Safety and security risks of large language models in healthcare

Article Title: Safety and security of large language models in healthcare

News Publication Date: 19 August 2026

Web References: https://doi.org/10.1038/s41586-026-10687-1

References: Clusmann J, Freyer O, Ostermann M, Ferber D, Ghaffari Laleh N, Hilgers L, Kolbinger FR, Schneider CV, Downing A, Wekenborg MK, Gilbert S, Foersch S, Truhn D, Wiest IC, Kather JN. “Safety and security of large language models in healthcare.” Nature. 2026. DOI: 10.1038/s41586-026-10687-1

Keywords: Large language models; artificial intelligence; healthcare; clinical AI; patient safety; cybersecurity; data privacy; prompt injection; data poisoning; hallucinations; automation bias; AI regulation; medical devices; AI governance

Leading the charge, Dr. Neal Chatterjee and his team at the University of Washington School of Medicine have harnessed the combined power of machine learning and clinical data to create predictive models that could potentially alter clinical practice. Published in the esteemed journal JACC: Advances, the research employed a vast dataset encompassing nearly 1.7 million patient records from a large integrated healthcare system in the U.S., encompassing both EHR data and 12-lead EKGs. The team’s approach leverages three distinct AI models: one informed solely by EKG waveforms, another utilizing structured EHR inputs comprising more than 150 clinical variables, and a third hybrid model integrating both data sources.

The methodology underpinning the model development was rigorous, stratified across three patient cohorts to ensure robustness and real-world applicability. Initially, the training cohort consisted of 993 out-of-hospital cardiac arrest cases alongside 5,479 age- and sex-matched control subjects without cardiac events, spanning nearly a decade from 2013 to 2021. This comprehensive dataset allowed the AI to discern subtle patterns and predictors embedded in both the electrical signatures of the heart and broader health parameters that correlate with increased SCA risk.

Validation proceeded with a testing cohort from more recent years (2022-2023), which included 463 cardiac arrest incidents and nearly 3,000 controls. Application of the AI models here confirmed their predictive fidelity, with the models reliably distinguishing high- and low-risk profiles congruent with training findings. However, the true test came from applying the tools to a real-world cohort: a large, unfiltered group of nearly 40,000 individuals who had undergone EKG testing in 2021 regardless of pre-existing conditions, followed longitudinally for two years to see who eventually suffered cardiac arrest.

Remarkably, the integrated EHR-EKG AI model correctly identified 153 of the 228 patients who experienced cardiac arrest as high-risk, exhibiting an enrichment in risk prediction that elevated from a baseline of 1 in 1,000 to 1 in 100. This degree of stratification could be transformative in clinical settings, alerting healthcare practitioners and patients alike to a risk magnitude impactful enough to prompt preemptive clinical decisions and potentially lifesaving interventions.

Notably, the EKG-only model – which depends solely on the analysis of the heart’s electrical activity – demonstrated impressive prognostic capability independently, showing only a modest decrease in performance compared to models incorporating the full range of EHR data. Given the global ubiquity and low cost of 12-lead EKG machines, this finding unlocks practical pathways for broad implementation of risk screening even outside advanced healthcare environments.

Beyond cardiovascular parameters traditionally associated with SCA, the AI models illuminated novel risk factors often overlooked in clinical practice. These included electrolyte imbalances, substance use behaviors, and adverse medication interactions, highlighting how multifaceted cardiac arrest triggers can be. This insight suggests that AI-driven risk alerts might encourage clinicians to systematically review modifiable patient factors and perform more nuanced, preventive care tailored to the individual’s comprehensive clinical profile.

Despite this promise, Dr. Chatterjee and his collaborators underscore that predictive power alone is insufficient without clear clinical pathways. The next frontier is refining post-prediction responses: determining which diagnostic tests, monitoring regimens, or therapeutic interventions should follow identification of elevated risk. Clarifying these management strategies is paramount to translating AI prediction into tangible reductions in SCA incidence and mortality.

Another caveat relates to the study’s data source—all drawn from a single healthcare system—raising questions about the generalizability of the models to demographically or geographically distinct populations. Additionally, the real-world cohort limitation to individuals who had undergone EKG testing introduces selection bias; patients not receiving EKGs, who might nonetheless be at risk, remain outside the model’s purview. Furthermore, concerns about AI model biases linked to healthcare disparities and demographic representation warrant careful ongoing evaluation to ensure equitable, unbiased application across diverse patient populations.

The research, funded by prestigious entities including the National Institutes of Health, the American Heart Association, the European Union, and the Foundation Leducq, represents a multi-institutional collaborative success involving Massachusetts General Hospital and the Broad Institute at MIT and Harvard. The confluence of clinical cardiology expertise, data science innovation, and vast patient data has created an unprecedented predictive toolset with the potential to radically change how sudden cardiac arrest is anticipated and perhaps eventually prevented.

Dr. Chatterjee points to an exciting era ahead where artificial intelligence transforms the interpretation of routine medical tests from static snapshots into dynamic, predictive analyses capable of forewarning life-threatening events. This evolution heralds a future in which the frustration and tragedy of sudden cardiac arrest—long an enigmatic killer striking without warning—may become significantly mitigated through enhanced data-driven foresight integrated seamlessly into everyday clinical workflows worldwide.

Subject of Research: People
Article Title: Artificial Intelligence-Enhanced Electrocardiography and Health Records to Predict Cardiac Arrest
News Publication Date: 11-May-2026
Web References: DOI: 10.1016/j.jacadv.2026.102787
Keywords: Cardiac arrest, Artificial intelligence, Electrocardiography, Electronic medical records, Computer modeling

Pathology reports have long served as the cornerstone for cancer diagnosis and treatment planning. However, as biomarker testing has proliferated and patient survival rates have improved, these reports have grown increasingly voluminous and intricate. Clinicians often face the challenge of sifting through multi-institutional, longitudinal data dense with histopathological, immunohistochemical, and genetic information, all under significant time constraints. Northwestern’s latest research addresses this critical bottleneck by deploying advanced large language models (LLMs) to generate succinct, comprehensive summaries that capture essential clinical details more reliably than physicians’ own written summaries.

The study’s authors meticulously analyzed 94 de-identified lung cancer pathology reports, encompassing a broad spectrum of diagnostic data including microscopic tumor characteristics, protein expression profiles, and molecular genetics that inform personalized treatment decisions. The team evaluated six open-source AI language models—Meta’s Llama 3.0, 3.1, and 3.2 variants, Google’s Gemma 9B, DeepSeek-R1, and Mistral 7.2B—each engineered to interpret and synthesize complex textual clinical data without reliance on external cloud-based chatbot frameworks.

Following model-generated summarization, a panel of expert oncologists rigorously assessed the outputs against physician-written clinical summaries. The consensus was striking: AI-generated summaries consistently outperformed their human counterparts, particularly in accurately incorporating molecular and genetic findings crucial for therapeutic strategies. The models’ ability to standardize and elevate the completeness of these summaries marks a significant milestone in addressing informational overload in oncology.

“The complexity of cancer care means clinicians must integrate ever-growing volumes of data, often under intense time pressures,” explained Dr. Mohamed Abazeed, senior study author and Chair of Radiation Oncology at Northwestern University Feinberg School of Medicine. “Our findings underscore that AI doesn’t replace clinical expertise but rather serves as a potent tool to ensure no critical pathological or genomic detail is overlooked—which can be a game-changer for patient outcomes.”

Not all AI architectures performed equally. DeepSeek and Meta’s Llama 3.1 models emerged as the strongest performers, demonstrating superior accuracy and completeness in summarization tasks. Importantly, these models are designed for local deployment, enabling hospital IT systems to integrate AI tools while maintaining patient data privacy—an increasingly vital consideration given heightened concerns about health information security.

Beyond accuracy, the potential clinical impact of this technology is profound. As Dr. Yirong Liu, lead author and radiation oncology resident at McGaw Medical Center, noted, “Patients with complex cancers undergo multiple biopsies and genetic tests across time. Their pathology reports often span dozens of pages. AI-driven summaries can spotlight elusive but critical information—like actionable genetic mutations—that might otherwise be missed, thereby enhancing treatment personalization and improving survival rates.”

The team is currently advancing this research by developing an application powered by Llama 3.1 which will enable clinicians to upload pathology reports and instantly receive AI-generated summaries for review. Nevertheless, the researchers emphasize that before such solutions enter routine clinical practice, extensive validation and testing across broader patient cohorts and cancer types are essential to establish reliability and safety.

This convergence of oncology and artificial intelligence represents a broader trend toward harnessing machine learning tools to manage clinical complexity and optimize workflow efficiency. Unlike conversational chatbots that generate generalized text, these AI systems are specifically trained to digest and condense exhaustive, technical reports into actionable clinical insights, thereby relieving physicians from repetitive, time-consuming documentation tasks.

The implications extend beyond lung cancer, with the potential to revolutionize pathology reporting in other cancer types and chronic diseases that require integrating multifaceted diagnostic data. By ensuring higher fidelity in the transmission of critical diagnostic information, AI-enabled summaries could become an indispensable support layer, augmenting clinical judgment and facilitating more informed decision-making pathways.

Funding for this pioneering work came from prestigious sources, including the Canadian Institute of Health Research and Amazon Web Services’ Social Impact program, reflecting the growing recognition of AI’s pivotal role in healthcare innovation. As these technologies mature, studies like Northwestern’s provide a foundational blueprint for developing AI-driven tools that prioritize patient safety, data security, and enhanced clinical usability.

The Northwestern Medicine study titled “Toward Automating the Summarization of Cancer Pathology Reports Using Large Language Models to Improve Clinical Usability” signals a transformative step forward. It illuminates a future where AI not only augments human intelligence but also fundamentally reshapes how vital medical knowledge is processed, delivered, and utilized in cancer care—potentially translating to better outcomes and improved quality of life for patients worldwide.

Subject of Research: Automating summarization of complex cancer pathology reports using large language models to improve clinical decision-making.

Article Title: Toward Automating the Summarization of Cancer Pathology Reports Using Large Language Models to Improve Clinical Usability

News Publication Date: April 8, 2026

Web References: DOI 10.1200/CCI-25-00284 (JCO Clinical Cancer Informatics)

References: Northwestern University study, JCO Clinical Cancer Informatics, April 8, 2026

Image Credits: Northwestern University