Background: Health care artificial intelligence (AI) systems are increasingly integrated into clinical workflows, yet remain vulnerable to data-poisoning attacks. A small number of manipulated training samples can compromise AI models used for diagnosis, documentation, and resource allocation. Existing privacy regulations, including the Health Insurance Portability and Accountability Act and the General Data Protection Regulation, may inadvertently complicate anomaly detection and cross-institutional auditing, thereby limiting visibility into adversarial activity. Objective: This study provides a comprehensive threat analysis of data poisoning vulnerabilities across major health care AI architectures. The goals are to (1) identify attack surfaces in clinical AI systems, (2) evaluate the feasibility and detectability of poisoning attacks analytically modeled in prior security research, and (3) propose a multilayered defense framework appropriate for health care settings. Methods: We synthesized empirical findings from 41 key security studies published between 2019 and 2025 and integrated them into an analytical threat-modeling framework specific to health care. We constructed 8 hypothetical yet technically grounded attack scenarios across 4 categories: (1) architecture-specific attacks on convolutional neural networks, large language models, and reinforcement learning agents (scenario A); (2) infrastructure exploitation in federated learning and clinical documentation pipelines (scenario B); (3) poisoning of critical resource allocation systems (scenario C); and (4) supply chain attacks affecting commercial foundation models (scenario D). Scenarios were aligned with realistic insider-access threat models and current clinical deployment practices. Results: Multiple empirical studies demonstrate that attackers with access to as few as 100-500 poisoned samples can compromise health care AI systems, with attack success rates typically ≥60%. Critically, attack success depends on the absolute number of poisoned samples rather than their proportion of the training corpus, a finding that fundamentally challenges assumptions that larger datasets provide inherent protection. We estimate that detection delays commonly range from 6 to 12 months and may extend to years in distributed or privacy-constrained environments. Analytical scenarios highlight that (1) routine insider access creates numerous injection points across health care data infrastructure, (2) federated learning amplifies risks by obscuring attribution, and (3) supply chain compromises can simultaneously affect dozens to hundreds of institutions. Privacy regulations further complicate cross-patient correlation and model audit processes, substantially delaying the detection of subtle poisoning campaigns. Conclusions: Health care AI systems face significant security challenges that current regulatory frameworks and validation practices do not adequately address. We propose a multilayered defense strategy that combines ensemble disagreement monitoring, adversarial testing, privacy-preserving yet auditable mechanisms, and strengthened governance requirements. Ensuring patient safety may require a shift from opaque, high-performance models toward more interpretable and constraint-driven architectures with verifiable robustness guarantees.

Jenny Liu is a solo first-time GP looking to back underrepresented wellness founders.

arXiv:2601.15059v1 Announce Type: new Abstract: Modern CI/CD pipelines integrating agent-generated code exhibit a structural failure in responsibility attribution. Decisions are executed through formally correct approval processes, yet no entity possesses both the authority to approve those decisions and the epistemic capacity to meaningfully understand their basis. We define this condition as responsibility vacuum: a state in which decisions occur, but responsibility cannot be attributed because authority and verification capacity do not coincide. We show that this is not a process deviation or technical defect, but a structural property of deployments where decision generation throughput exceeds bounded human verification capacity. We identify a scaling limit under standard deployment assumptions, including parallel agent generation, CI-based validation, and individualized human approval gates. Beyond a throughput threshold, verification ceases to function as a decision criterion and is replaced by ritualized approval based on proxy signals. Personalized responsibility becomes structurally unattainable in this regime. We further characterize a CI amplification dynamic, whereby increasing automated validation coverage raises proxy signal density without restoring human capacity. Under fixed time and attention constraints, this accelerates cognitive offloading in the broad sense and widens the gap between formal approval and epistemic understanding. Additional automation therefore amplifies, rather than mitigates, the responsibility vacuum. We conclude that unless organizations explicitly redesign decision boundaries or reassign responsibility away from individual decisions toward batch- or system-level ownership, responsibility vacuum remains an invisible but persistent failure mode in scaled agent deployments.

arXiv:2601.14305v1 Announce Type: cross Abstract: The increase in the number of Internet of Things (IoT) devices has tremendously increased the attack surface of cyber threats thus making a strong intrusion detection system (IDS) with a clear explanation of the process essential towards resource-constrained environments. Nevertheless, current IoT IDS systems are usually traded off with detection quality, model elucidability, and computational effectiveness, thus the deployment on IoT devices. The present paper counteracts these difficulties by suggesting an explainable AI (XAI) framework based on an optimized Decision Tree classifier with both local and global importance methods: SHAP values that estimate feature attribution using local explanations, and Morris sensitivity analysis that identifies the feature importance in a global view. The proposed system attains the state of art on the test performance with 99.91% accuracy, F1-score of 99.51% and Cohen Kappa of 0.9960 and high stability is confirmed by a cross validation mean accuracy of 98.93%. Efficiency is also enhanced in terms of computations to provide faster inferences compared to those that are generalized in ensemble models. SrcMac has shown as the most significant predictor in feature analyses according to SHAP and Morris methods. Compared to the previous work, our solution eliminates its major drawback lack because it allows us to apply it to edge devices and, therefore, achieve real-time processing, adhere to the new regulation of transparency in AI, and achieve high detection rates on attacks of dissimilar classes. This combination performance of high accuracy, explainability, and low computation make the framework useful and reliable as a resource-constrained IoT security problem in real environments.

arXiv:2601.14525v1 Announce Type: cross Abstract: Automated AI research holds great potential to accelerate scientific discovery. However, current LLMs often generate plausible-looking but ineffective ideas. Execution grounding may help, but it is unclear whether automated execution is feasible and whether LLMs can learn from the execution feedback. To investigate these, we first build an automated executor to implement ideas and launch large-scale parallel GPU experiments to verify their effectiveness. We then convert two realistic research problems - LLM pre-training and post-training - into execution environments and demonstrate that our automated executor can implement a large fraction of the ideas sampled from frontier LLMs. We analyze two methods to learn from the execution feedback: evolutionary search and reinforcement learning. Execution-guided evolutionary search is sample-efficient: it finds a method that significantly outperforms the GRPO baseline (69.4% vs 48.0%) on post-training, and finds a pre-training recipe that outperforms the nanoGPT baseline (19.7 minutes vs 35.9 minutes) on pre-training, all within just ten search epochs. Frontier LLMs often generate meaningful algorithmic ideas during search, but they tend to saturate early and only occasionally exhibit scaling trends. Reinforcement learning from execution reward, on the other hand, suffers from mode collapse. It successfully improves the average reward of the ideator model but not the upper-bound, due to models converging on simple ideas. We thoroughly analyze the executed ideas and training dynamics to facilitate future efforts towards execution-grounded automated AI research.

arXiv:2601.15161v1 Announce Type: cross Abstract: Large Language Models (LLMs) are increasingly used for clinical decision support, where hallucinations and unsafe suggestions may pose direct risks to patient safety. These risks are particularly challenging as they often manifest as subtle clinical errors that evade detection by generic metrics, while expert-authored fine-grained rubrics remain costly to construct and difficult to scale. In this paper, we propose a retrieval-augmented multi-agent framework designed to automate the generation of instance-specific evaluation rubrics. Our approach grounds evaluation in authoritative medical evidence by decomposing retrieved content into atomic facts and synthesizing them with user interaction constraints to form verifiable, fine-grained evaluation criteria. Evaluated on HealthBench, our framework achieves a Clinical Intent Alignment (CIA) score of 60.12%, a statistically significant improvement over the GPT-4o baseline (55.16%). In discriminative tests, our rubrics yield a mean score delta ($\mu_{\Delta} = 8.658$) and an AUROC of 0.977, nearly doubling the quality separation achieved by GPT-4o baseline (4.972). Beyond evaluation, our rubrics effectively guide response refinement, improving quality by 9.2% (from 59.0% to 68.2%). This provides a scalable and transparent foundation for both evaluating and improving medical LLMs. The code is available at https://anonymous.4open.science/r/Automated-Rubric-Generation-AF3C/.

arXiv:2505.20310v2 Announce Type: replace Abstract: Meta-analysis is a systematic research methodology that synthesizes data from multiple existing studies to derive comprehensive conclusions. This approach not only mitigates limitations inherent in individual studies but also facilitates novel discoveries through integrated data analysis. Traditional meta-analysis involves a complex multi-stage pipeline including literature retrieval, paper screening, and data extraction, which demands substantial human effort and time. However, while LLM-based methods can accelerate certain stages, they still face significant challenges, such as hallucinations in paper screening and data extraction. In this paper, we propose a multi-agent system, Manalyzer, which achieves end-to-end automated meta-analysis through tool calls. The hybrid review, hierarchical extraction, self-proving, and feedback checking strategies implemented in Manalyzer significantly alleviate these two hallucinations. To comprehensively evaluate the performance of meta-analysis, we construct a new benchmark comprising 729 papers across 3 domains, encompassing text, image, and table modalities, with over 10,000 data points. Extensive experiments demonstrate that Manalyzer achieves significant performance improvements over the LLM baseline in multi meta-analysis tasks. Project page: https://black-yt.github.io/meta-analysis-page/ .

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

GigaTIME leverages multimodal AI to generate virtual multiplex immunofluorescence (mIF) profiles from standard H&E slides, enabling comprehensive tumor immune microenvironment modeling across a large (>14,000) and diverse patient population. This virtual approach unlocks new opportunities for large-scale clinical discoveries that were previously hindered by the scarcity of mIF data.

OpenEvidence, maker of a popular chatbot that helps doctors search clinical evidence, on Wednesday announced $250 million in new funding.

The new round led by Thrive Capital and DST Global values OpenEvidence at $12 billion, and the company has announced $735 million in funding in the last 12 months. OpenEvidence is free to use by any clinician with a national provider identifier number. The company’s primary business model is advertising shown to clinicians. 

Founded in 2022, OpenEvidence is one of the most prominent and best-funded companies from a wave of health artificial intelligence companies that emerged since the widespread availability of large language models. Reflecting on the eye-popping fundraising for health AI, a Silicon Valley Bank report released earlier in January raised an eyebrow at the ability of companies like OpenEvidence to deliver on their stratospheric valuations with advertising and software-as-a-service business models. “It won’t be a surprise to see them tap the value of the data they’re already collecting” to offer more services to pharma or other customers, the authors wrote.

Continue to STAT+ to read the full story…

© Adobe

arXiv:2601.13122v1 Announce Type: new Abstract: Modern general-purpose AI systems made using large language and vision models, are capable of performing a range of tasks like writing text articles, generating and debugging codes, querying databases, and translating from one language to another, which has made them quite popular across industries. However, there are risks like hallucinations, toxicity, and stereotypes in their output that make them untrustworthy. We review various risks and vulnerabilities of modern general-purpose AI along eight widely accepted responsible AI (RAI) principles (fairness, privacy, explainability, robustness, safety, truthfulness, governance, and sustainability) and compare how they are non-existent or less severe and easily mitigable in traditional task-specific counterparts. We argue that this is due to the non-deterministically high Degree of Freedom in output (DoFo) of general-purpose AI (unlike the deterministically constant or low DoFo of traditional task-specific AI systems), and there is a need to rethink our approach to RAI for general-purpose AI. Following this, we derive C2V2 (Control, Consistency, Value, Veracity) desiderata to meet the RAI requirements for future general-purpose AI systems, and discuss how recent efforts in AI alignment, retrieval-augmented generation, reasoning enhancements, etc. fare along one or more of the desiderata. We believe that the goal of developing responsible general-purpose AI can be achieved by formally modeling application- or domain-dependent RAI requirements along C2V2 dimensions, and taking a system design approach to suitably combine various techniques to meet the desiderata.

arXiv:2601.13846v1 Announce Type: new Abstract: This paper introduces Virtual Urbanism (VU), a multimodal AI-driven analytical framework for quantifying urban identity through the medium of synthetic urban replicas. The framework aims to advance computationally tractable urban identity metrics. To demonstrate feasibility, the pilot study Virtual Urbanism and Tokyo Microcosms is presented. A pipeline integrating Stable Diffusion and LoRA models was used to produce synthetic replicas of nine Tokyo areas rendered as dynamic synthetic urban sequences, excluding existing orientation markers to elicit core identity-forming elements. Human-evaluation experiments (I) assessed perceptual legitimacy of replicas; (II) quantified area-level identity; (III) derived core identity-forming elements. Results showed a mean identification accuracy of ~81%, confirming the validity of the replicas. Urban Identity Level (UIL) metric enabled assessment of identity levels across areas, while semantic analysis revealed culturally embedded typologies as core identity-forming elements, positioning VU as a viable framework for AI-augmented urban analysis, outlining a path toward automated, multi-parameter identity metrics.

arXiv:2601.11544v1 Announce Type: cross Abstract: The introduction of large language models (LLMs) has greatly enhanced the capabilities of software agents. Instead of relying on rule-based interactions, agents can now interact in flexible ways akin to humans. However, this flexibility quickly becomes a problem in fields where errors can be disastrous, such as in a pharmacy context, but the opposite also holds true; a system that is too inflexible will also lead to errors, as it can become too rigid to handle situations that are not accounted for. Work using LLMs in a pharmacy context have adopted a wide scope, accounting for many different medications in brief interactions -- our strategy is the opposite: focus on a more narrow and long task. This not only enables a greater understanding of the task at hand, but also provides insight into what challenges are present in an interaction of longer nature. The main challenge, however, remains the same for a narrow and wide system: it needs to strike a balance between adherence to conversational requirements and flexibility. In an effort to strike such a balance, we present a prototype system meant to provide medication counseling while juggling these two extremes. We also cover our design in constructing such a system, with a focus on methods aiming to fulfill conversation requirements, reduce hallucinations and promote high-quality responses. The methods used have the potential to increase the determinism of the system, while simultaneously not removing the dynamic conversational abilities granted by the usage of LLMs. However, a great deal of work remains ahead, and the development of this kind of system needs to involve continuous testing and a human-in-the-loop. It should also be evaluated outside of commonly used benchmarks for LLMs, as these do not adequately capture the complexities of this kind of conversational system.

arXiv:2601.11567v1 Announce Type: cross Abstract: Small open-source medical large language models (LLMs) offer promising opportunities for low-resource deployment and broader accessibility. However, their evaluation is often limited to accuracy on medical multiple choice question (MCQ) benchmarks, and lacks evaluation of consistency, robustness, or reasoning behavior. We use MCQ coupled to human evaluation and clinical review to assess six small open-source medical LLMs (HuatuoGPT-o1 (Chen 2024), Diabetica-7B, Diabetica-o1 (Wei 2024), Meditron3-8B (Sallinen2025), MedFound-7B (Liu 2025), and ClinicaGPT-base-zh (Wang 2023)) in pediatric endocrinology. In deterministic settings, we examine the effect of prompt variation on models' output and self-assessment bias. In stochastic settings, we evaluate output variability and investigate the relationship between consistency and correctness. HuatuoGPT-o1-8B achieved the highest performance. The results show that high consistency across the model response is not an indicator of correctness, although HuatuoGPT-o1-8B showed the highest consistency rate. When tasked with selecting correct reasoning, both HuatuoGPT-o1-8B and Diabetica-o1 exhibit self-assessment bias and dependency on the order of the candidate explanations. Expert review of incorrect reasoning rationales identified a mix of clinically acceptable responses and clinical oversight. We further show that system-level perturbations, such as differences in CUDA builds, can yield statistically significant shifts in model output despite stable accuracy. This work demonstrates that small, semantically negligible prompt perturbations lead to divergent outputs, raising concerns about reproducibility of LLM-based evaluations and highlights the output variability under different stochastic regimes, emphasizing the need of a broader diagnostic framework to understand potential pitfalls in real-world clinical decision support scenarios.

arXiv:2601.12471v1 Announce Type: cross Abstract: Current evaluation of large language models (LLMs) overwhelmingly prioritizes accuracy; however, in real-world and safety-critical applications, the ability to abstain when uncertain is equally vital for trustworthy deployment. We introduce MedAbstain, a unified benchmark and evaluation protocol for abstention in medical multiple-choice question answering (MCQA) -- a discrete-choice setting that generalizes to agentic action selection -- integrating conformal prediction, adversarial question perturbations, and explicit abstention options. Our systematic evaluation of both open- and closed-source LLMs reveals that even state-of-the-art, high-accuracy models often fail to abstain with uncertain. Notably, providing explicit abstention options consistently increases model uncertainty and safer abstention, far more than input perturbations, while scaling model size or advanced prompting brings little improvement. These findings highlight the central role of abstention mechanisms for trustworthy LLM deployment and offer practical guidance for improving safety in high-stakes applications.

arXiv:2601.12607v1 Announce Type: cross Abstract: As Large Language Models (LLMs) become ubiquitous across various scientific domains, their lack of ability to perform complex tasks like running simulations or to make complex decisions limits their utility. LLM-based agents bridge this gap due to their ability to call external resources and tools and thus are now rapidly gaining popularity. However, coming up with a workflow that can balance the models, cloud providers, and external resources is very challenging, making implementing an agentic system more of a hindrance than a help. In this work, we present a domain-agnostic, model-independent workflow for an agentic framework that can act as a scientific assistant while being run entirely on cloud. Built with a supervisor agent marshaling an array of agents with individual capabilities, our framework brings together straightforward tasks like literature review and data analysis with more complex ones like simulation runs. We describe the framework here in full, including a proof-of-concept system we built to accelerate the study of Catalysts, which is highly important in the field of Chemistry and Material Science. We report the cost to operate and use this framework, including the breakdown of the cost by services use. We also evaluate our system on a custom-curated synthetic benchmark and a popular Chemistry benchmark, and also perform expert validation of the system. The results show that our system is able to route the task to the correct agent 90% of the time and successfully complete the assigned task 97.5% of the time for the synthetic tasks and 91% of the time for real-world tasks, while still achieving better or comparable accuracy to most frontier models, showing that this is a viable framework for other scientific domains to replicate.

arXiv:2601.12946v1 Announce Type: cross Abstract: Generative artificial intelligence (AI) is rapidly populating medical records with synthetic content, creating a feedback loop where future models are increasingly at risk of training on uncurated AI-generated data. However, the clinical consequences of this AI-generated data contamination remain unexplored. Here, we show that in the absence of mandatory human verification, this self-referential cycle drives a rapid erosion of pathological variability and diagnostic reliability. By analysing more than 800,000 synthetic data points across clinical text generation, vision-language reporting, and medical image synthesis, we find that models progressively converge toward generic phenotypes regardless of the model architecture. Specifically, rare but critical findings, including pneumothorax and effusions, vanish from the synthetic content generated by AI models, while demographic representations skew heavily toward middle-aged male phenotypes. Crucially, this degradation is masked by false diagnostic confidence; models continue to issue reassuring reports while failing to detect life-threatening pathology, with false reassurance rates tripling to 40%. Blinded physician evaluation confirms that this decoupling of confidence and accuracy renders AI-generated documentation clinically useless after just two generations. We systematically evaluate three mitigation strategies, finding that while synthetic volume scaling fails to prevent collapse, mixing real data with quality-aware filtering effectively preserves diversity. Ultimately, our results suggest that without policy-mandated human oversight, the deployment of generative AI threatens to degrade the very healthcare data ecosystems it relies upon.

arXiv:2601.13580v1 Announce Type: cross Abstract: We introduce Neural Organ Transplantation (NOT), a modular adaptation framework that enables trained transformer layers to function as reusable transferable checkpoints for domain adaptation. Unlike conventional fine-tuning approaches that tightly couple trained parameters to specific model instances and training data, NOT extracts contiguous layer subsets ("donor organs") from pre-trained models, trains them independently on domain-specific data, and saves them as standalone checkpoint files that can be transplanted into compatible recipient models without access to the original training data. Through experiments on three decoder-only transformer architectures spanning 124M to 20B parameters (GPT-2, TinyLlama, and GPT-OSS), we demonstrate that donor transplantation substantially outperforms existing adaptation methods, achieving an order-of-magnitude improvement in perplexity over LoRA while training significantly faster. The method exhibits position dependence, with early insertion positions yielding optimal results. Cross-domain transfer at billion-parameter scale reveals unexpected regularization benefits. These findings demonstrate that transformer middle layers can support efficient modular transfer for decoder-only architectures, enabling privacy-preserving expertise sharing through checkpoint distribution. We note that this approach is currently limited to decoder-only models; preliminary experiments on encoder-based architectures show reduced effectiveness.

arXiv:2601.13710v1 Announce Type: cross Abstract: Artificial intelligence has reshaped medical imaging, yet the use of AI on clinical data for prospective decision support remains limited. We study pre-operative prediction of clinically meaningful improvement in chronic rhinosinusitis (CRS), defining success as a more than 8.9-point reduction in SNOT-22 at 6 months (MCID). In a prospectively collected cohort where all patients underwent surgery, we ask whether models using only pre-operative clinical data could have identified those who would have poor outcomes, i.e. those who should have avoided surgery. We benchmark supervised ML (logistic regression, tree ensembles, and an in-house MLP) against generative AI (ChatGPT, Claude, Gemini, Perplexity), giving each the same structured inputs and constraining outputs to binary recommendations with confidence. Our best ML model (MLP) achieves 85 % accuracy with superior calibration and decision-curve net benefit. GenAI models underperform on discrimination and calibration across zero-shot setting. Notably, GenAI justifications align with clinician heuristics and the MLP's feature importance, repeatedly highlighting baseline SNOT-22, CT/endoscopy severity, polyp phenotype, and physchology/pain comorbidities. We provide a reproducible tabular-to-GenAI evaluation protocol and subgroup analyses. Findings support an ML-first, GenAI- augmented workflow: deploy calibrated ML for primary triage of surgical candidacy, with GenAI as an explainer to enhance transparency and shared decision-making.