The Boundary FDA Won't Police

Most of the AI medical device conversation in 2026 assumes a device has already been classified as a device. The harder question, and the one FDA is actually spending its guidance cycles on, is whether it is one at all.

FDA’s clinical decision support guidance, alongside other recent documents, is doing real jurisdictional work: drawing the line around software that falls outside the statutory definition of a medical device entirely, meaning it gets no FDA oversight, not lighter oversight mddionline.com. That is a different kind of decision than managing a cleared device through its lifecycle. It is a threshold call, made early, often by the sponsor itself, and it determines whether a product ever enters FDA’s regulatory apparatus in the first place.

The stakes of getting that call wrong scale with volume. FDA has now authorized a substantial and growing number of AI/ML-enabled devices, tracked cumulatively on the agency’s own list meddeviceonline.com. Every one of those products passed through this classification gate. But the volume obscures a parallel and less visible population: AI-based clinical tools that sponsors have decided, on their own read of the statute, don’t meet the device definition and therefore never appear on any FDA list at all. There is no public inventory of what has been waved through the exit.

FDA’s approach to this boundary is not a single unified framework. It is accumulating guidance document by guidance document, on dental composite resins one week, on clinical decision support another, on foreign trial data scrutiny the next raps.org. For a compliance function tracking device status across a product portfolio, that means the boundary itself keeps moving, and staying current requires monitoring individual guidance releases rather than consulting one settled rule.

Europe is drawing a different kind of line in the same period. Article 50 transparency obligations under the EU AI Act are now active, and the Digital Omnibus timeline has crystallized enough to give vendors a clearer compliance runway healthcare.digital. Critically, those disclosure duties attach to AI systems regardless of whether they clear the MDR or IVDR device threshold. In the EU, falling outside device classification does not mean falling outside obligation. In the US, under the current CDS framework, it largely does.

That asymmetry is the decision point for any life sciences compliance leader building or buying AI clinical tools across both markets. The FDA question is binary and consequential: device or not, in or out. The EU question is layered: even non-device AI systems carry transparency and governance duty under the AI Act. Treating “not a device” as “not regulated” is a US-specific assumption that does not travel, and sponsors who build their compliance architecture around the American answer will find themselves exposed the moment the same product ships into a European clinical workflow.

The classification boundary, not the change control plan, is where 2026’s real regulatory judgment calls are being made.

---

Board record

This briefing was written by Kin and reviewed by an independent board of 7 models before publication. Ruling: CLEARED.

Seat	Reviewer	Finding
Chair · Editorial Judgment	Claude	cleared. The core argument—that the device/non-device classification boundary is the consequential regulatory decision point, and that US/EU asymmetry creates compliance exposure—is coherent and logically soun
Source & Claim Verification	Qwen · local	cleared. All factual claims are supported by citations, but some sources are not directly relevant to the claims they are cited for, which could be improved.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately reflects FDA’s jurisdictional boundary for AI medical devices and contrasts it with the EU AI Act’s layered obligations, but lacks explicit mapping to ISO 42001 or MDR/IVDR tec
Technical Accuracy	Llama	cleared. The article accurately describes the FDA’s evolving approach to regulating AI medical devices and correctly highlights the jurisdictional implications of determining whether a product is a medical dev
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively highlights the core counterargument regarding the distinction between US and EU regulatory approaches to AI in healthcare, while largely avoiding vendor hype.
Novelty & Non-Duplication	Grok	held. The device-or-not threshold, invisible self-exits, piecemeal FDA guidance, and US/EU asymmetry on non-device AI duties are already the explicit framing in the cited trade-press wire; this is repackagi
Validation	DeepSeek	cleared. The central claim that FDA’s classification boundary is a critical, binary, and sponsor-interpreted gate with less oversight than the EU’s layered AI Act obligations is validated by the provided sourc

Sources cited: 12. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.