OpenEvidence exits Europe: what AI Act uncertainty means for clinical AI

TL;DR

OpenEvidence has withdrawn from the EU and UK, citing mounting regulatory uncertainty around the EU AI Act — leaving European physicians without clinically grounded AI at the point of care. Meanwhile, the US clinical AI market accelerated sharply through April, with major funding rounds, new platform expansions, and free access for verified clinicians. The implications for pharma now divide sharply by geography. The AI Act was designed to raise the safety floor for medical AI in Europe; its first observable effect has been to push physicians toward less-grounded alternatives.

OpenEvidence withdrew from the European Union and the UK, citing "mounting regulatory uncertainty regarding the treatment of AI systems," including the EU AI Act.

40 percent of US physicians use OpenEvidence daily across more than 10,000 hospitals. The withdrawal makes the geography of clinical AI access official: it is American.

The contrast with the past three weeks of US activity is stark.

• April 3: OpenEvidence + Tandem expansion into prescription generation and prior authorization inside the platform.
• Mid-April: Abridge raised a $316 million Series E extension on top of its $5.3 billion valuation, with NEJM and JAMA partnerships moving the product from ambient documentation into clinical decision support.
• April 22: OpenAI launched ChatGPT for Clinicians, free for verified US clinicians.

US physicians are getting clinically grounded AI inside the workflow. European physicians get general-purpose LLMs without medical grounding or clinical filtering, and even ambient scribes are being adopted at a glacial pace.

Three months ago I argued OpenEvidence had the strongest wedge in clinical AI. April confirmed it. The implication for pharma now divides by geography.

What this means for US pharma

For US pharma, the transition is accelerating. The $20–25 billion US pharma digital ad market has a new high-CPM channel, with CPMs of $70 to over $1,000 versus $5 to $15 on social media. MSL teams face structural pressure. KOL strategy becomes content licensing strategy. Sales reps and journal advertising are being supplemented and then displaced by an AI mediator at the point of care.

What this means for European pharma

For European pharma, the channel does not exist. The traditional rep model survives somewhat by default (if the physician actually lets the rep into their office or hospital).

European physicians wanting clinical AI have ChatGPT, Claude, or Gemini without peer-reviewed grounding. This is a worse patient safety outcome than the AI Act intended.

Three markers to watch in the next 18 months

Three markers to watch in the next 18 months. European clinically-grounded AI emerging with AI Act compliance built in. European pharma engaging the AI Act high-risk classification to enable medical AI deployment. European physicians migrating to general-purpose LLMs in clinical workflow despite the absence of clinical filtering.

The AI Act was designed to raise the safety floor for medical AI in Europe. Its first observable effect on the clinical AI market has been to push physicians toward less-grounded alternatives. Regulatory unintended consequences rarely show up this fast or this visibly. This could be worse than the cookie banner regime, version two.

Key takeaways

• OpenEvidence's withdrawal from the EU and UK makes the geography of clinical AI access officially American — 40 percent of US physicians use it daily across more than 10,000 hospitals.
• April saw a sharp acceleration in US clinical AI: platform expansion into prescription generation and prior authorization, a $316 million Abridge Series E extension, and free ChatGPT access for verified US clinicians.
• For US pharma, a new high-CPM channel is emerging at the point of care, with CPMs of $70 to over $1,000 versus $5 to $15 on social media — structurally pressuring MSL teams, KOL strategy, and the traditional sales rep model.
• For European pharma, that channel does not exist, and the traditional rep model survives somewhat by default.
• European physicians wanting clinical AI are left with general-purpose LLMs without peer-reviewed grounding — a worse patient safety outcome than the AI Act intended.
• The AI Act's first observable effect on the clinical AI market has been to push physicians toward less-grounded alternatives, with regulatory unintended consequences showing up unusually fast and visibly.
• I'm watching three markers over the next 18 months: EU-compliant clinically grounded AI emerging, European pharma engaging the high-risk classification to enable deployment, and European physicians migrating to general-purpose LLMs in clinical workflow.