A Top Clinical AI Left Europe. The FDA Eased Its US Path

OpenEvidence stopped serving doctors in the UK and EU on April 30, citing regulatory uncertainty under the EU AI Act. The company is valued at $12 billion after a January Series D led by Thrive Capital and DST Global, with backing from GV, Sequoia, NVIDIA, Kleiner Perkins, and Mayo Clinic. Its clinical search tool is used daily by more than 40% of US physicians across 10,000 hospitals, processing roughly 18 million clinical consultations in December 2025 alone.

The EU AI Act treats clinical AI tools as high-risk and requires conformity assessment, documented training data, bias testing, post-market monitoring, and explainability, on top of existing Medical Device Regulation obligations. The Digital Omnibus trilogue on April 28 broke down over how those overlapping obligations should work together, reopened a week later, and reached a provisional agreement on May 7. Medical devices were the one category left out of the simplification. Industrial machinery moved into a lighter conformity track. AI in regulated medical software stayed under both the AI Act and the Medical Device Regulation. The category OpenEvidence operates in was the line Europe chose not to relax.

Three months earlier, the United States went the other direction. On January 6, 2026, the FDA issued new clinical decision support guidance, which expands the carveout in Section 3060 of the 21st Century Cures Act, which excludes clinical decision support software from medical device regulation if the tool meets four criteria, including that it supports rather than replaces clinician judgment and that a clinician can independently review and understand the basis for the recommendations. Arnold & Porter’s analysis of the guidance describes it as aligned with administration efforts to reduce regulatory barriers to AI.

OpenEvidence is the kind of tool the expanded carveout was designed to protect from device classification. The tool surfaces evidence to physicians. It does not acquire medical signals, interpret images, or direct treatment, and physicians retain decision authority over the output. Whether the product formally satisfies the four-criteria test is a determination for the company and the FDA, but every public-facing feature of the tool fits the carveout’s design. The FDA device pathway would not require the kind of public conformity assessment, bias audit, training-data documentation, post-market monitoring, or patient disclosure that the European framework is moving toward. The company could publish more of that information voluntarily. Based on public materials, it has not published the equivalent of a European-style conformity record.

The two regulators moved in opposite directions on the same category of tool in the same four-month window. Europe spent the spring working out how to evaluate medical AI under two overlapping legal regimes, and decided the dual conformity burden was severe enough to require its own trilogue cycle. The United States spent January loosening the only regime it had, on the reasoning that the existing rules were slowing AI in clinical settings. A company sitting between those postures has an obvious choice. OpenEvidence made it.

The FDA’s reading of the Cures Act has its defenders. The exclusion is the law Congress wrote. The 2026 guidance is the kind of clarifying update agencies routinely issue, and reasonable people argue for keeping evidence-summary tools outside the medical device pathway. The US has built one regime for clinical AI. That regime turns on whether a tool replaces or supports physician judgment. Tools on the support side of the line have no other public-accountability mechanism. A clinical AI used in 18 million encounters per month sits inside the exclusion. There is no second regime underneath.

European clinicians are not without alternatives. Tools including iatroX, Praxis Medicine, Medwise AI, and AMBOSS AI Mode have moved into the gap, built on national guideline systems and designed to operate under European requirements. The vendor with the largest US footprint and the highest valuation in the category is not among them. American patients use the tool those vendors are building alternatives to, under a regime that does not require it to be evaluated the way the European regime would.

Deliberate exclusion is not the same as delay. The Cures Act framework was chosen on purpose. The question the framework leaves open is whether a tool that helps shape 18 million clinical decisions a month should have any public accountability beyond the choice not to regulate it as a device.

OpenEvidence may yet find a path back to Europe. There are reports that predict the company will eventually navigate compliance given its resources. If that happens, European patients will once again have access under a regime that asks the company to document what the tool does and how well it does it. American patients will continue to have access without ever having asked the question.