arXiv:2602.05183v2 Announce Type: replace-cross Abstract: Large language models (LLMs) are increasingly trained in complex Reinforcement Learning, multi-agent environments, making it difficult to understand how behavior changes over training. Sparse Autoencoders (SAEs) have recently shown to be useful for data-centric interpretability. In this work, we analyze large-scale reinforcement learning training runs from the sophisticated environment of Full-Press Diplomacy by applying pretrained SAEs, alongside LLM-summarizer methods. We introduce Meta-Autointerp, a method for grouping SAE features into interpretable hypotheses about training dynamics. We discover fine-grained behaviors including role-playing patterns, degenerate outputs, language switching, alongside high-level strategic behaviors and environment-specific bugs. Through automated evaluation, we validate that 90% of discovered SAE Meta-Features are significant, and find a surprising reward hacking behavior. However, through two user studies, we find that even subjectively interesting and seemingly helpful SAE features may be worse than useless to humans, along with most LLM generated hypotheses. However, a subset of SAE-derived hypotheses are predictively useful for downstream tasks. We further provide validation by augmenting an untrained agent's system prompt, improving the score by +14.2%. Overall, we show that SAEs and LLM-summarizer provide complementary views into agent behavior, and together our framework forms a practical starting point for future data-centric interpretability work on ensuring trustworthy LLM behavior throughout training.

Nature Biotechnology, Published online: 09 February 2026; doi:10.1038/s41587-026-03003-9

Insights into regulatory T cell biology are accelerating therapeutic innovation in cancer immunotherapy, autoimmune diseases and transplant rejection.

Extrachromosomal DNA (ecDNA) is a major source of oncogenic fusions across cancer types, generating tissue-specific fusion landscapes with diagnostic potential. EcDNA-borne PVT1 5′-end fusions stabilize partner RNAs and boost oncogene output.

arXiv:2601.09117v1 Announce Type: cross Abstract: Generative AI systems increasingly enable the production of highly realistic synthetic media. Civitai, a popular community-driven platform for AI-generated content, operates a monetized feature called Bounties, which allows users to commission the generation of content in exchange for payment. To examine how this mechanism is used and what content it incentivizes, we conduct a longitudinal analysis of all publicly available bounty requests collected over a 14-month period following the platform's launch. We find that the bounty marketplace is dominated by tools that let users steer AI models toward content they were not trained to generate. At the same time, requests for content that is "Not Safe For Work" are widespread and have increased steadily over time, now comprising a majority of all bounties. Participation in bounty creation is uneven, with 20% of requesters accounting for roughly half of requests. Requests for "deepfake" - media depicting identifiable real individuals - exhibit a higher concentration than other types of bounties. A nontrivial subset of these requests involves explicit deepfakes despite platform policies prohibiting such content. These bounties disproportionately target female celebrities, revealing a pronounced gender asymmetry in social harm. Together, these findings show how monetized, community-driven generative AI platforms can produce gendered harms, raising questions about consent, governance, and enforcement.

This is the online version of The Readout, STAT’s flagship biotech newsletter. Sign up to get it in your inbox.

You’re back. We’re sort of back. It’s Day 2 of JPM and we’re definitely not exhausted or delirious yet.

This is Elaine Chen, Adam Feuerstein, Matt Herper, and Allison DeAngelis again. We’ve got a lot more news today, so let’s get to it.

The next test for Kite Pharma — and Gilead

It’s anito-cel, the CAR-T therapy for multiple myeloma that Gilead is developing in partnership with Arcellx. Gilead submitted the therapy to the FDA sometime before the end of December, Cindy Perettie, executive vice president of Kite Pharma, the cell therapy division of Gilead, told STAT at a Gilead media breakfast.

Continue to STAT+ to read the full story…

© Alex Hogan/STAT

Nature Medicine, Published online: 09 January 2026; doi:10.1038/s41591-025-04170-z

Analysis of a placebo-controlled trial of a BCMA-targeting CAR-T cell therapy in patients with myasthenia gravis shows that CAR-T cell infusion selectively remodels the systemic immune environment, with elimination of BCMA-high plasma cells and activated plasmacytoid dendritic cells and changes in the autoreactive B cell repertoire.

Background: Structured medication reviews (SMRs) are an essential component of medication optimization, especially for patients with multimorbidity and polypharmacy. However, the process remains challenging due to the complexities of patient data, time constraints, and the need for coordination among health care professionals (HCPs). This study explores HCPs’ perspectives on the integration of artificial intelligence (AI)–assisted tools to enhance the SMR process, with a focus on the potential benefits of and barriers to adoption. Objective: This study aims to identify the key user requirements for AI-assisted tools to improve the efficiency and effectiveness of SMRs, specifically for patients with multimorbidity, complex polypharmacy, and frailty. Methods: A qualitative study was conducted involving focus groups and semistructured interviews with HCPs and patients in the United Kingdom. Participants included physicians, pharmacists, clinical pharmacologists, psychiatrists from primary and secondary care, a policy maker, and patients with multimorbidity. Data were analyzed using a hybrid inductive and deductive thematic analysis approach to identify themes related to AI-assisted tool functionality, workflow integration, user-interface visualization, and usability in the SMR process. Results: Four major themes emerged from the analysis: innovative AI potential, optimizing electronic patient record visualization, functionality of the AI tool for SMRs, and facilitators of and barriers to AI tool implementation. HCPs identified the potential of AI to support patient identification and prioritizing those at risk of medication-related harm. AI-assisted tools were viewed as essential in detecting prescribing gaps, drug interactions, and patient risk trajectories over time. Participants emphasized the importance of presenting patient data in an intuitive format, with a patient interface for shared decision-making. Suggestions included color-coding blood results, highlighting critical medication reviews, and providing timelines of patient medical histories. HCPs stressed the need for AI tools to integrate seamlessly with existing electronic patient record systems and provide actionable insights without overwhelming users with excessive notifications or “pop-up” alerts. Factors influencing the uptake of AI-assisted tools included the need for user-friendly design, evidence of tool effectiveness (though some were skeptical about the predictive accuracy of AI models), and addressing concerns around digital exclusion. Conclusions: The findings highlight the potential for AI-assisted tools to streamline and optimize the SMR process, particularly for patients with multimorbidity and complex polypharmacy. However, successful implementation depends on addressing concerns related to workflow integration, user acceptance, and evidence of effectiveness. User-centered design is crucial to ensure that AI-assisted tools support HCPs in delivering high-quality, patient-centered care while minimizing cognitive overload and alert fatigue.

arXiv:2512.22568v1 Announce Type: new Abstract: The phenomenal advances in large language models (LLMs) and other foundation models over the past few years have been based on optimizing large-scale transformer models on the surprisingly simple objective of minimizing next-token prediction loss, a form of predictive coding that is also the backbone of an increasingly popular model of brain function in neuroscience and cognitive science. However, current foundation models ignore three other important components of state-of-the-art predictive coding models: tight integration of actions with generative models, hierarchical compositional structure, and episodic memory. We propose that to achieve safe, interpretable, energy-efficient, and human-like AI, foundation models should integrate actions, at multiple scales of abstraction, with a compositional generative architecture and episodic memory. We present recent evidence from neuroscience and cognitive science on the importance of each of these components. We describe how the addition of these missing components to foundation models could help address some of their current deficiencies: hallucinations and superficial understanding of concepts due to lack of grounding, a missing sense of agency/responsibility due to lack of control, threats to safety and trustworthiness due to lack of interpretability, and energy inefficiency. We compare our proposal to current trends, such as adding chain-of-thought (CoT) reasoning and retrieval-augmented generation (RAG) to foundation models, and discuss new ways of augmenting these models with brain-inspired components. We conclude by arguing that a rekindling of the historically fruitful exchange of ideas between brain science and AI will help pave the way towards safe and interpretable human-centered AI.

arXiv:2511.12439v2 Announce Type: replace Abstract: Online health resources and large language models (LLMs) are increasingly used as a first point of contact for medical decision-making, yet their reliability in healthcare remains limited by low accuracy, lack of transparency, and susceptibility to unverified information. We introduce a proof-of-concept conversational self-triage system that guides LLMs with 100 clinically validated flowcharts from the American Medical Association, providing a structured and auditable framework for patient decision support. The system leverages a multi-agent framework consisting of a retrieval agent, a decision agent, and a chat agent to identify the most relevant flowchart, interpret patient responses, and deliver personalized, patient-friendly recommendations, respectively. Performance was evaluated at scale using synthetic datasets of simulated conversations. The system achieved 95.29% top-3 accuracy in flowchart retrieval (N=2,000) and 99.10% accuracy in flowchart navigation across varied conversational styles and conditions (N=37,200). By combining the flexibility of free-text interaction with the rigor of standardized clinical protocols, this approach demonstrates the feasibility of transparent, accurate, and generalizable AI-assisted self-triage, with potential to support informed patient decision-making while improving healthcare resource utilization.

arXiv:2512.01241v2 Announce Type: replace-cross Abstract: Large language models (LLMs) are routinely used by physicians and patients for medical advice, yet their clinical safety profiles remain poorly characterized. We present NOHARM (Numerous Options Harm Assessment for Risk in Medicine), a benchmark using 100 real primary care-to-specialist consultation cases to measure frequency and severity of harm from LLM-generated medical recommendations. NOHARM covers 10 specialties, with 12,747 expert annotations for 4,249 clinical management options. Across 31 LLMs, potential for severe harm from LLM recommendations occurs in up to 22.2% (95% CI 21.6-22.8%) of cases, with harm of omission accounting for 76.6% (95% CI 76.4-76.8%) of errors. Safety performance is only moderately correlated (r = 0.61-0.64) with existing AI and medical knowledge benchmarks. The best models outperform generalist physicians on safety (mean difference 9.7%, 95% CI 7.0-12.5%), and a diverse multi-agent approach improves safety compared to solo models (mean difference 8.0%, 95% CI 4.0-12.1%). Therefore, despite strong performance on existing evaluations, widely used AI models can produce severely harmful medical advice at nontrivial rates, underscoring clinical safety as a distinct performance dimension necessitating explicit measurement.

arXiv:2512.13724v1 Announce Type: cross Abstract: Neurological diseases are the leading global cause of disability, yet most lack disease-modifying treatments. We present PROTON, a heterogeneous graph transformer that generates testable hypotheses across molecular, organoid, and clinical systems. To evaluate PROTON, we apply it to Parkinson's disease (PD), bipolar disorder (BD), and Alzheimer's disease (AD). In PD, PROTON linked genetic risk loci to genes essential for dopaminergic neuron survival and predicted pesticides toxic to patient-derived neurons, including the insecticide endosulfan, which ranked within the top 1.29% of predictions. In silico screens performed by PROTON reproduced six genome-wide $\alpha$-synuclein experiments, including a split-ubiquitin yeast two-hybrid system (normalized enrichment score [NES] = 2.30, FDR-adjusted $p

arXiv:2512.12500v1 Announce Type: cross Abstract: Artificial intelligence (AI) is increasingly permeating healthcare, from physician assistants to consumer applications. Since AI algorithm's opacity challenges human interaction, explainable AI (XAI) addresses this by providing AI decision-making insight, but evidence suggests XAI can paradoxically induce over-reliance or bias. We present results from two large-scale experiments (623 lay people; 153 primary care physicians, PCPs) combining a fairness-based diagnosis AI model and different XAI explanations to examine how XAI assistance, particularly multimodal large language models (LLMs), influences diagnostic performance. AI assistance balanced across skin tones improved accuracy and reduced diagnostic disparities. However, LLM explanations yielded divergent effects: lay users showed higher automation bias - accuracy boosted when AI was correct, reduced when AI erred - while experienced PCPs remained resilient, benefiting irrespective of AI accuracy. Presenting AI suggestions first also led to worse outcomes when the AI was incorrect for both groups. These findings highlight XAI's varying impact based on expertise and timing, underscoring LLMs as a "double-edged sword" in medical AI and informing future human-AI collaborative system design.

arXiv:2510.02967v3 Announce Type: replace-cross Abstract: This paper presents the development and evaluation of a Retrieval-Augmented Generation (RAG) system for querying the United Kingdom's National Institute for Health and Care Excellence (NICE) clinical guidelines using Large Language Models (LLMs). The extensive length and volume of these guidelines can impede their utilisation within a time-constrained healthcare system, a challenge this project addresses through the creation of a system capable of providing users with precisely matched information in response to natural language queries. The system's retrieval architecture, composed of a hybrid embedding mechanism, was evaluated against a corpus of 10,195 text chunks derived from three hundred guidelines. It demonstrates high performance, with a Mean Reciprocal Rank (MRR) of 0.814, a Recall of 81% at the first chunk and of 99.1% within the top ten retrieved chunks, when evaluated on 7901 queries. The most significant impact of the RAG system was observed during the generation phase. When evaluated on a manually curated dataset of seventy question-answer pairs, RAG-enhanced models showed substantial gains in performance. Faithfulness, the measure of whether an answer is supported by the source text, was increased by 64.7 percentage points to 99.5% for the RAG-enhanced O4-Mini model and significantly outperformed the medical-focused Meditron3-8B LLM, which scored 43%. Clinical evaluation by seven Subject Matter Experts (SMEs) further validated these findings, with GPT-4.1 achieving 98.7% accuracy while reducing unsafe responses by 67% compared to O4-Mini (from 3.0 to 1.0 per evaluator). This study thus establishes RAG as an effective, reliable, and scalable approach for applying generative AI in healthcare, enabling cost-effective access to medical guidelines.

Background: Increasing demand on healthcare systems requires innovative and transformative solutions to deliver efficient, high-quality care. One promising approach is Digital Twin (DT) technology, which leverages real time data to create dynamic virtual representations of a physical entity (individuals or space) to anticipate future scenarios and support care decisions. While DTs have been explored in various sectors, their application in Hospital at Home (HaH), which delivers acute level care in home environments, remains unexplored. Objective: This review bridges a critical knowledge gap and examines the existing evidence on DT-enabling tools for managing patients with frailty in home settings. This will identify the underpinning architectural components required to inform a HaH-DT system which can support clinical decision-making. Methods: Six electronic databases (Embase, MEDLINE, Cochrane CENTRAL, CINAHL, Web of Science and Scopus) were searched, along with grey literature, to identifying primary studies published in English, between January 2019 and September 2025. Included studies had to report on the monitoring or management of patients with frailty within their own home, and information was charted on a pre-defined data collection form to answer the research objectives. Review articles, protocols, and conference abstracts were excluded. Results: Sixty-nine reports were included, of which 54% (n=37) used quantitative approaches, and 36% (n=25) were pilot or feasibility studies. Reports were analysed for DT-enabling tools and systematically mapped across the proposed five-layered DT architecture: sensing, communication, storage, analytics, and visualisation. Taxonomies of DT layers, their interconnections, and the classifications of the types of data collected (e.g., about the patient, the home environment, the use of medical equipment) are presented. This evidence identifies DT-enabling tools used for a variety of functions and a range of sensing technologies that exist (e.g., passive sensing via wearables, active physiological sensors, ambient sensors to detect motion/environmental changes). The most prevalent modes of communication were wireless and network-based (n=36), with the majority using Bluetooth (n=12). This review highlights better understanding of data management, in particular secure storage, is required within local healthcare systems. The emerging potential of predictive and prescriptive analytics, which can enable clinicians to predict risk, support clinical decision-making, or activate alert-triggered health interventions were mapped. Existing evidence suggests analytics methods are currently largely descriptive with a lack of advanced methods such as prescriptive analytics to enable recommendations of an optimal course of action, and the absence of diagnostic analytics which can highlight why a situation has occurred. Reported DT-enabling tools demonstrate patient-centered benefits, including enhanced motivation, reassurance, and personalised care. However, concerns persist regarding device accuracy, user acceptability, and implications for carers and organisational workflows. Conclusions: This review is among the first to systematically map DT-enabling tools to inform a potential HaH-DT in patients with frailty and organised by a 5-layered conceptual model. Understanding these architectural layers provides the foundations to enable stakeholders advance research and development in areas where there are knowledge gaps and consider how a HaH DT can effectively operate within current healthcare systems. By leveraging technology-enabled care in complex home-based settings, there is great potential to deliver safer, personalised and timely care.

Nature, Published online: 10 December 2025; doi:10.1038/s41586-025-09792-4

High-depth sequencing of non-cancerous tissue from patients with metastatic cancer reveals single-base mutational signatures of alcohol, smoking and cancer treatments, and reveals how exogenous factors, including cancer therapies, affect somatic cell evolution.

arXiv:2509.26153v3 Announce Type: replace Abstract: Large language models (LLMs) integrated into agent-driven workflows hold immense promise for healthcare, yet a significant gap exists between their potential and practical implementation within clinical settings. To address this, we present a practitioner-oriented field manual for deploying generative agents that use electronic health record (EHR) data. This guide is informed by our experience deploying the "irAE-Agent", an automated system to detect immune-related adverse events from clinical notes at Mass General Brigham, and by structured interviews with 21 clinicians, engineers, and informatics leaders involved in the project. Our analysis reveals a critical misalignment in clinical AI development: less than 20% of our effort was dedicated to prompt engineering and model development, while over 80% was consumed by the sociotechnical work of implementation. We distill this effort into five "heavy lifts": data integration, model validation, ensuring economic value, managing system drift, and governance. By providing actionable solutions for each of these challenges, this field manual shifts the focus from algorithmic development to the essential infrastructure and implementation work required to bridge the "valley of death" and successfully translate generative AI from pilot projects into routine clinical care.

arXiv:2511.20680v1 Announce Type: cross Abstract: Despite high performance on clinical benchmarks, large language models may reach correct conclusions through faulty reasoning, a failure mode with safety implications for oncology decision support that is not captured by accuracy-based evaluation. In this two-cohort retrospective study, we developed a hierarchical taxonomy of reasoning errors from GPT-4 chain-of-thought responses to real oncology notes and tested its clinical relevance. Using breast and pancreatic cancer notes from the CORAL dataset, we annotated 600 reasoning traces to define a three-tier taxonomy mapping computational failures to cognitive bias frameworks. We validated the taxonomy on 822 responses from prostate cancer consult notes spanning localized through metastatic disease, simulating extraction, analysis, and clinical recommendation tasks. Reasoning errors occurred in 23 percent of interpretations and dominated overall errors, with confirmation bias and anchoring bias most common. Reasoning failures were associated with guideline-discordant and potentially harmful recommendations, particularly in advanced disease management. Automated evaluators using state-of-the-art language models detected error presence but could not reliably classify subtypes. These findings show that large language models may provide fluent but clinically unsafe recommendations when reasoning is flawed. The taxonomy provides a generalizable framework for evaluating and improving reasoning fidelity before clinical deployment.

arXiv:2511.19482v2 Announce Type: replace-cross Abstract: This study investigates how human experts evaluate the capacity of Generative AI (GenAI) to contextualize STEAM education in the Global South, with a focus on Ghana. Using a convergent mixed-methods design, four STEAM specialists assessed GenAI-generated lesson plans created with a customized Culturally Responsive Lesson Planner (CRLP) and compared them to standardized lesson plans from the Ghana National Council for Curriculum and Assessment (NaCCA). Quantitative ratings were based on a validated 25-item Culturally Responsive Pedagogy Rubric measuring bias awareness, cultural representation, contextual relevance, linguistic responsiveness, and teacher agency. Qualitative reflections provided additional insight into how GenAI handles cultural and pedagogical appropriateness. Findings show that GenAI, when paired with the CRLP tool, can support contextualized STEAM instruction by linking abstract curriculum standards to learners' cultural knowledge, community practices, and everyday experiences. Experts rated GenAI-assisted lessons as more culturally grounded and pedagogically responsive than NaCCA plans, integrating Indigenous knowledge, bilingual elements, and locally relevant examples. However, GenAI struggled to represent Ghana's cultural pluralism, often offering surface-level references to language, history, and identity. These weaknesses were most evident in Mathematics and Computing, where cultural nuance was limited. The results highlight the need for continued teacher mediation, community involvement, and culturally attuned refinement of AI outputs. Future work should include classroom trials, expanded expert participation, and model fine-tuning using Indigenous language corpora to strengthen cultural fidelity in Global South contexts.

arXiv:2511.12439v1 Announce Type: new Abstract: Online health resources and large language models (LLMs) are increasingly used as a first point of contact for medical decision-making, yet their reliability in healthcare remains limited by low accuracy, lack of transparency, and susceptibility to unverified information. We introduce a proof-of-concept conversational self-triage system that guides LLMs with 100 clinically validated flowcharts from the American Medical Association, providing a structured and auditable framework for patient decision support. The system leverages a multi-agent framework consisting of a retrieval agent, a decision agent, and a chat agent to identify the most relevant flowchart, interpret patient responses, and deliver personalized, patient-friendly recommendations, respectively. Performance was evaluated at scale using synthetic datasets of simulated conversations. The system achieved 95.29% top-3 accuracy in flowchart retrieval (N=2,000) and 99.10% accuracy in flowchart navigation across varied conversational styles and conditions (N=37,200). By combining the flexibility of free-text interaction with the rigor of standardized clinical protocols, this approach demonstrates the feasibility of transparent, accurate, and generalizable AI-assisted self-triage, with potential to support informed patient decision-making while improving healthcare resource utilization.