Industry
OpenEvidence: The AI That Took US Medicine by Storm — and Why It Left the EU
OpenEvidence is the most talked-about clinical AI in US medicine right now — and, as of spring 2026, it is no longer available to clinicians in the EU or UK. Both facts are true, and the gap between them is one of the most instructive stories in healthcare AI. This piece explains what OpenEvidence actually is, why it took the US by storm, and — carefully, because the internet is already full of confident misinformation about it — what really happened with Europe.
First, an important clarification for readers of this site: OpenEvidence is not an AI medical scribe. It does not transcribe encounters or write your notes. It's a clinical answer engine — a cited medical-evidence search tool clinicians query at the point of care. We don't rank it among the AI scribes because it's a different product category. We're covering it because it's the most important adjacent tool to the scribe market, and because its European exit is a preview of pressures the whole sector faces.
(This piece is the what-happened explainer. For our verdict on the product itself — what OpenEvidence does well, where it's risky, and who should rely on it — see our separate OpenEvidence review.)
What OpenEvidence actually is
Founded in 2022 by Daniel Nadler (founder of Kensho) and Zachary Ziegler, and incubated through the Mayo Clinic Platform Accelerate program, OpenEvidence answers free-text clinical questions conversationally, synthesising the literature with inline citations to primary sources. It has multi-year content agreements with the NEJM Group (Feb 2025) and the JAMA Network (Jun 2025), plus content relationships reported with the AMA and NCCN.
Its model is unusual: free for verified clinicians (identity confirmed via US National Provider Identifier or hospital email), monetised by advertising, with the company stating a separation between ads and clinical answers. It positions hard on being citation-grounded — teaching the reasoning behind each answer and, per analyst write-ups, declining to answer when it can't properly source a claim. (That's the company's positioning relayed by reputable analysts, not an independent audit — the same scepticism we apply to scribe accuracy claims in our hallucinations guide applies here.)
Why it took the US by storm
The traction is genuinely exceptional, and the figures are consistently reported:
- The CEO told CNBC (Jan 2026) that ~40% of US physicians use OpenEvidence daily, across 10,000+ hospitals; JAMA's mid-2025 announcement put it at "over a third of doctors in the United States."
- Company-reported usage grew from ~430,000 registered physicians and 8.5M monthly consultations (Jul 2025) to ~760,000 and ~18M (Dec 2025).
- A funding escalator that's hard to overstate: Series A (~$1B, Feb 2025) → Series B ($210M at $3.5B, Jul 2025) → Series C ($200M at ~$6B, Oct 2025) → Series D ($250M at a $12B valuation, Jan 2026), with Sequoia, GV, Kleiner Perkins, Thrive, DST, Blackstone, Nvidia and the Mayo Clinic among backers.
- It reported scoring 100% on the USMLE in 2025.
Three things explain the storm: it's free at the point of use for clinicians, it's citation-grounded enough to be trusted in the moment, and it landed exactly as physicians became comfortable asking an AI clinical questions. It has also begun surfacing its evidence content inside ambient workflows — including a collaboration to appear within Microsoft Dragon Copilot — which is why scribe buyers increasingly encounter it.
What actually happened in the EU and UK
Here is where accuracy matters, and where most coverage gets it wrong. As of late April 2026, OpenEvidence is no longer available to clinicians in the EU and UK. An in-app notice (first surfaced publicly via the HIStalk news feed on 28 April 2026 and then echoed across health-tech trade press) reportedly attributed the move to "mounting regulatory uncertainty regarding the treatment of AI systems in the European Union and the United Kingdom, including … the EU Artificial Intelligence Act."
Two corrections to the common narrative:
- It was a voluntary withdrawal, not a ban. No EU body and no UK regulator (e.g., the MHRA) is reported to have prohibited the product. Every credible source frames this as a company decision that the legal risk outweighed near-term EU/UK commercial opportunity. Anyone telling you OpenEvidence was "banned" or "made illegal" in Europe is overstating what's documented.
- Even before withdrawal, it was effectively US-gated. Verification was built around the US NPI, with no real pathway for EU/UK clinicians — so practical European availability was always thin.
(Sourcing caveat, in keeping with how this site works: the verbatim wording of that notice traces to a screenshot reported by trade outlets, not a first-party OpenEvidence document we could verify. We state the fact of withdrawal and the reported reason, and label the rest as reported-not-confirmed.)
The deeper "why": clinical-decision software in the EU
The user-facing reason cited the AI Act, but the structural reason any tool that informs clinical decisions struggles in the EU runs deeper — and it's worth understanding because it affects scribes too.
Under the EU Medical Device Regulation (Regulation (EU) 2017/745, MDR) and the Commission's qualification guidance MDCG 2019‑11, standalone software can be a regulated medical device when it has a medical purpose of its own. The decisive classification rule is MDR Annex VIII, Rule 11: software intended to provide information used to take decisions with diagnostic or therapeutic purposes is classified at minimum Class IIa, rising to IIb or III as the potential for patient harm increases. A simple search/retrieval tool may stay outside the device definition; a tool that synthesises evidence to inform diagnosis or therapy is squarely in the zone Rule 11 was written for.
Being a Class IIa+ device under MDR is not a paperwork formality. It triggers Notified Body conformity assessment, CE marking, a clinical evaluation, technical documentation and ongoing post-market surveillance — and, where AI is involved, overlapping obligations under the EU AI Act, whose high-risk timelines were still shifting through 2026. That is a fundamentally heavier path than the US, where the 21st Century Cures Act carves out "non-device" clinical decision support when the software lets the clinician independently review the basis for its recommendations (the FDA narrowed, then in 2026 partly re-loosened, this line).
The net: the EU's default posture treats decision-informing clinical software as a regulated medical device; the US default carves out an exemption hinged on transparency and clinician independent review. A US-born, free-to-clinician, ad-supported answer engine is far easier to scale under the second regime than the first — which is the real backdrop to OpenEvidence's "regulatory uncertainty" exit, whatever the precise trigger.
Why this matters for AI scribes
Pure ambient scribes that only transcribe and draft a note for a clinician to review sit in a lighter regulatory position than a tool that suggests diagnoses or therapy — but the line moves the moment a scribe starts adding clinical suggestions, coding, or "impressions." It's why several European scribes we cover lead with CE marking and ISO 13485 (e.g. Haidy / 44ai, Tandem Health) and why data-residency and medical-device status are first-class criteria in our methodology and our compliance buyer's checklist. OpenEvidence's exit is the clearest signal yet that "we'll just launch the US product in Europe later" is not a safe assumption in clinical AI.
Bottom line
OpenEvidence is a genuinely landmark US clinical AI — citation-grounded, free to clinicians, used daily by a huge share of US physicians, and valued at $12B. It is also, as of spring 2026, withdrawn from the EU and UK by its own decision, citing regulatory uncertainty — not banned by a regulator. The distinction matters, and so does the lesson: in European healthcare AI, regulatory fit is not an afterthought you bolt on later. For the tools we do rank, that's exactly why we treat compliance and device status as core, not cosmetic — see the rankings and methodology.