healthcare

The paper introduces BABA-Bench, a benchmark of 47 interactive patient-advocacy worlds that checks each agent action in real time for permissions, evidence, and prohibited steps. Evaluating 14 frontier models across 1,974 runs, it finds agents often reach correct outcomes via invalid routes—e.g., Claude Opus 5 hits all outcomes in 81.6% of runs but passes all procedural checks only 52.5%—and that supplying a verified route boosts pass rates to over 91%.

The paper introduces Noura, a framework that turns executable personal clinical specifications into supervision for LLM-based diet planning, enabling compact models like Qwen3-14B to dramatically improve multi-day meal plan compliance and reduce rule violations compared to direct generation by larger models.

A Nature paper introduces EVLF-FM, an 8B-parameter clinical multimodal vision-language foundation model with fine-grain explainability that outperforms clinicians in radiology, dermatology and other medical tasks, showing frontier-level medical intelligence can run without data-center scale compute.

This paper introduces the concept of 'disparity laundering,' showing that clinical multi-agent LLM teams do not eliminate demographic bias but instead carry it through into consensus decisions, with interventions only working when counterfactual versions are made visible and pooled during deliberation.

Bill Gates notes that Alzheimer's breakthroughs are advancing faster than the healthcare systems meant to deliver them, leaving many patients without access, and argues that closing this gap is the next major frontier.

Elon Musk announced that Neuralink's first product, named Telepathy, enables high-speed communication with computers or humans through thought alone, with output bandwidth expected to far exceed human speech or typing in coming years.

AI agents powered by Claude Sonnet 5 reportedly performed brain surgery on a mouse, which recovered afterward, per Exin Therapeutics. The result suggests growing potential for autonomous AI-driven surgical robotics.

Anthropic used its Claude model in partnership with Divergence Labs to generate around 35,000 antibody designs in weeks, with some candidates outperforming Xolair at dislodging cell-bound IgE in lab tests, accelerating food allergy treatment research.

A person reportedly followed ChatGPT's advice to reduce their seizure medication, resulting in devastating health consequences, highlighting the risks of AI systems dispensing medical guidance.

Anthropic announced deep research partnerships with six organizations developing new medicines, and has begun its own drug discovery efforts, sharing early results on how Claude can support drug development stages.

OpenRadView is a free, open-source desktop viewer for personal CT, MRI and X-ray scans, supporting DICOM and NIfTI files with linked axial/coronal/sagittal views, experimental 3D and measurements, plus optional MCP access for AI agents.

约翰斯·霍普金斯湾景医疗中心通知约3,182名患者，其个人健康信息因一名员工将表格上传至未经批准的公共AI工具而可能泄露，医院正加强政策与员工培训。

Researchers developed and deployed LLENS, an on-premises clinical support LLM agent platform running inside Hokkaido University Hospital's closed network using the open-weight Kimi K2.6 model on 8 NVIDIA H200 GPUs, achieving 97.2% accuracy on hospital document questions via interactive tool-based access compared to 81.3% with fixed retrieval.

Biograph announced the first U.S. study using Caristo Diagnostics' AI-powered CaRi-Heart technology, finding that 42.9% of asymptomatic, low-risk adults had coronary inflammation — often undetectable by standard blood tests, cholesterol markers, or traditional cardiovascular risk scores.

InclusionAI (Ant Group) announced Ling-3.0-flash-Sante, a 124B-parameter (5.1B active) healthcare-focused model with a 256K context window for medical reasoning, literature research, and tool-assisted workflows, with weights released on Hugging Face.

蚂蚁集团 inclusionAI 发布医疗专用模型 Ling-3.0-flash-Sante，基于 Ling-3.0-flash 构建，采用 124B 总参数/5.1B 激活参数的混合线性 MoE 架构，支持 256K 长上下文，面向医疗推理、深度研究和工具辅助工作流，权重已在 Hugging Face 与 ModelScope 开放。

Hugging Face released Carbon-X, an open DNA language model that answers questions about gene sequences in plain English, scoring 66.7% on DNA benchmarks near GPT-6.1 Sol at roughly 1/1,800 the cost per problem. It runs on a single 80GB GPU, potentially making large-scale variant classification dramatically cheaper for hospitals and researchers, though it is not yet clinically validated.

Hugging Face released Carbon-X, an open model that lets users chat with DNA in natural language, trained on 300 billion tokens of genomic sequences and English text to help researchers interpret genetic mutations and design therapies.

NVIDIA 和其欧洲合作伙伴宣布多项利用开放模型的公共服务项目，涵盖欧盟委员会的 AI@EC 平台、波兰的医疗影像与就业服务智能体系统，以及英国本地化数据建设。

A research paper presenting an LLM-based pipeline that automates psychotherapy fidelity assessment and generates actionable supervisory feedback from clinical transcripts, evaluated with open-weight models and validated on an RCT dataset with mental health expert user studies.