News updates
Hand-curated by AIH editors. Every item carries a link to the primary source so you can read the original.
Governor Polis signed SB 189 on May 14, 2026, narrowing the Colorado AI Act's original consequential-decision regime and delaying enforcement from June 30, 2026 to January 1, 2027. Healthcare use cases remain in scope but with materially reduced documentation, impact-assessment, and developer-disclosure obligations.
The editors argue medicine has always been an information enterprise, and that LLMs mark a qualitative shift from retrieval to synthesis. Efficiency gains are real, but the editorial flags new reliability and accountability risks that clinicians and journals will need to absorb — not delegate.
McDermott and colleagues review the Medical Event Data Standard, an open framework aimed at the long-standing reproducibility and interoperability gap in EHR-based AI. The review reports adoption across 21 institutions and 27+ papers as of March 2026.
Darshak Sanghavi argues the most consequential AI deployment in U.S. healthcare today is not at the bedside — it is the escalating arms race between hospital revenue-cycle AI and insurer program-integrity AI, a fight he estimates burns more than $140B/year while patients absorb the collateral damage.
Tempus AI is open-sourcing the Paige Image Management System — slide viewer (FullFocus), case management (FullFolio), and AI orchestration modules — under a new consortium with YNHH and MSK as founding academic members, aiming for a common pathology platform that competing AI vendors can build on.
The WHO's new discussion paper, 'Artificial Intelligence and Evidence-Informed Policy', maps where AI helps and where it harms across the policy cycle — from defining problems to monitoring outcomes — and argues that AI should augment, not replace, human judgement in Member State decision-making.
The agency's running catalog of cleared AI/ML-enabled devices keeps growing. The list is the authoritative public record of which AI tools have crossed the U.S. regulatory bar — even if most authorizations remain concentrated in imaging.
The World Health Organization's standing guidance on AI for health — including a 2024 update specifically addressing large multi-modal models — remains the most-cited international framework for ethics, safety, and governance of generative AI tools deployed against clinical use cases.
CHAI's framework sets concrete expectations for evaluating and monitoring health-AI tools — covering bias testing, performance drift, and use-case documentation — without certifying products. U.S. health systems are increasingly citing it as the baseline a candidate AI tool should meet before clinical deployment.