Health AI Week: Conferences, Partnerships, and Regulation Converge

Health AI had a packed week — an ethics conference opened in Baltimore, Johns Hopkins partnered with OpenEvidence on human-first clinical AI, the federal government launched four ARPA-H programs to speed clinical trials, the FDA's comment window on generative AI medical devices closes Sunday, and twenty years of drug discovery innovation converged in London.

A National Summit on Ethical AI in Public Health Opens in Baltimore

The National Network of Public Health Institutes (NNPHI) kicked off AIxPH 2026 on October 14 in Baltimore, bringing together public health practitioners, technologists, ethicists, and policymakers for a three-day conversation about how AI is reshaping population-level health practice.

The conference is organized around four pillars: AI governance and accountability, disease surveillance and outbreak detection, community and equity-centered health applications, and privacy protections in data-intensive systems. Those pillars reflect real tensions that have emerged as public health agencies begin deploying predictive tools — questions about who controls models, how surveillance data can be used, and whether AI-driven resource allocation reinforces or corrects historical inequities.

AIxPH is one of the first convenings designed specifically for public health professionals rather than hospital administrators or biomedical researchers. Epidemiology and community health operate on different data types, accountability structures, and risk tolerances than clinical medicine, and the governance frameworks being shaped at the conference reflect those differences.

OpenEvidence and Johns Hopkins Medicine Put Humans at the Center of Clinical AI

On October 13, OpenEvidence and Johns Hopkins Medicine announced a collaboration aimed at integrating AI into both clinical decision-making and medical education — with a deliberate emphasis on keeping clinicians in the loop.

The partnership combines OpenEvidence's evidence synthesis platform with Johns Hopkins clinical protocols, patient history tools, and institutional best practices. The stated goal is not to replace physician judgment but to surface relevant evidence faster, reduce cognitive load during high-stakes decisions, and flag when a patient's situation diverges from standard guidance.

What makes the collaboration notable is its dual scope. It is simultaneously trying to change how practicing clinicians work today and how medical students learn to work tomorrow. Training programs that don't include AI workflows are preparing students for a clinical environment that is already changing. By embedding AI tools in education alongside clinical practice, Johns Hopkins and OpenEvidence are betting that the most durable path to good outcomes is training clinicians who are both skeptical and competent users of AI systems.

HHS Bets on AI to Break the Clinical Trials Bottleneck

The U.S. Department of Health and Human Services, through the Advanced Research Projects Agency for Health (ARPA-H), announced four AI programs targeting the clinical trials pipeline — a domain where average development costs exceed $2 billion per drug and failure rates top 90 percent.

The four initiatives span distinct pressure points in the trial lifecycle:

SURPASS focuses on trial design, using AI to model outcomes and optimize protocols before a single patient is enrolled.

STACK addresses data infrastructure, building computational systems that make trial data more accessible and interoperable across sites and sponsors.

COMMONS broadens patient enrollment by matching eligible participants to trials they would otherwise never encounter — a problem that disproportionately affects rural and underserved communities.

CINCH targets oncology specifically, using AI to match cancer patients to suitable trials based on molecular profiles, treatment histories, and eligibility criteria.

Together the programs reflect ARPA-H's approach to health R&D: rather than funding individual therapies, it funds the infrastructure and methods that determine whether any therapy can be tested efficiently. Clinical trials are the single largest bottleneck between scientific discovery and patient access, and AI-enabled improvements to their speed and inclusiveness could have compounding effects across medicine.

FDA Closes Public Comment on Generative AI Medical Devices This Sunday

The Food and Drug Administration's August 2026 discussion paper on regulating generative AI-based medical devices reaches its public comment deadline on October 19. For anyone working in medical device development, clinical informatics, or health AI policy, the window is closing.

The paper focuses on what makes generative AI different from previous software-as-a-medical-device (SaMD) categories. Traditional SaMD models are typically locked after training: regulators can evaluate a specific version, and that version doesn't change. Generative AI models can adapt through continued learning, produce novel outputs that weren't explicitly anticipated during development, and behave differently across patient populations in ways that are harder to characterize in premarket testing.

The FDA is asking commenters to help define what premarket frameworks should look like for systems that evolve, how post-market surveillance should work when a model's outputs can shift over time, and how to handle AI systems that combine generative components with traditional diagnostic tools. With the comment period closing this week, the agency is moving toward proposed rulemaking — making the current window unusually high-stakes for anyone with a stake in how these rules get written.

Twenty Years of Drug Discovery Innovation, Now AI-Enabled, in London

ELRIG's Drug Discovery conference marked its twentieth anniversary on October 14–15 at London's ExCeL centre, and AI was woven throughout the agenda rather than siloed into a dedicated track — a signal of how thoroughly the technology has been absorbed into modern pharmaceutical R&D.

The conference brings together researchers across next-generation biophysics, multi-omics, microbiology, and translational science. This year AI-enabled discovery appeared at each stage of the pipeline: in target identification using large-scale genomic data, in molecule design and optimization, in assay development, and in the translation of preclinical findings to human trials.

Two decades of drug discovery conferences have tracked the field's passage through combinatorial chemistry, high-throughput screening, and computational docking — and now machine learning at scale. The twenty-year milestone is a useful frame for appreciating how much has changed. The fundamental challenge of identifying which interventions are safe and effective in humans remains as hard as ever. What AI is changing is the rate at which hypotheses can be tested and the range of chemical and biological space that can be explored.