The Readiness Gap Behind FDA's 1,000 AI Devices

The Readiness Gap Behind FDA’s 1,000 AI Devices

FDA has authorized more than 1,000 AI-enabled devices to date, a number regulatory teams inside medtech companies cite as proof they already know how to work with this agency on artificial intelligence. Healthcare Dive’s tally, cited in a recent analysis of the FDA’s October docket, comes with an important qualifier: the overwhelming majority of that install base is not generative AI www.marketscale.com. Most of it is locked or narrowly adaptive machine learning, cleared under a regulatory paradigm built for models that don’t change behavior at inference time. That distinction is about to matter a great deal.

FDA’s discussion paper on generative AI-enabled devices, open for public comment, asks manufacturers and clinicians to help define how the agency should treat systems that generate novel outputs rather than classify against fixed categories www.digitaljournal.com. The agency has floated Predetermined Change Control Plans as one mechanism for managing this, extending a framework it already uses for adaptive but bounded algorithms www.bipc.com. But PCCPs were designed around defined modification protocols and known performance boundaries. Generative and agentic systems, by construction, produce outputs the manufacturer did not fully specify in advance. The compliance muscle memory built over a decade of locked-model submissions, heavy on data lineage documentation and static validation studies, does not automatically transfer to that problem www.meddeviceonline.com.

The practical risk is complacency dressed up as experience. A regulatory affairs team that has cleared a dozen AI-enabled devices under existing pathways may reasonably believe its submission playbook scales. It does not, at least not without material rework. The marketscale analysis frames the shift bluntly: FDA’s current posture rewards teams that can produce test evidence over teams that can produce polished narrative decks [www.marketscale.com](https://www.marketscale.com/industries/healthcare/fdas-oct-19-ai

---

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 experience with locked ML models doesn’t transfer to generative AI regulation—is coherent and logically sound, but the claim that PCCPs are inadequate for generative systems res
Source & Claim Verification	Qwen · local	cleared. Most factual claims are supported by citations, but a few lines lack specific references, such as the claim about the majority of AI devices not being generative AI.
Regulatory & Framework Fidelity	Mistral	cleared. The briefing accurately reflects FDA’s evolving stance on generative AI but does not substantively address ISO 42001, EU AI Act, or MDR/IVDR requirements.
Technical Accuracy	Llama	cleared. The article accurately highlights the distinction between locked/narrowly adaptive machine learning models and generative AI, and the regulatory challenges posed by the latter, but could benefit from
Bias, Balance & Hype Control	Gemini	cleared. The briefing effectively identifies and counters potential vendor hype by distinguishing between existing AI device approvals and the emerging challenges of generative AI regulation, explicitly statin
Novelty & Non-Duplication	Grok	held. The briefing largely repackages widely circulated wire items on FDA’s gen-AI discussion paper, the 1,000-device tally, and PCCP limits, adding only a thin ‘readiness gap/complacency’ frame already pre
Validation	DeepSeek	cleared. The central claim that most of the FDA’s 1,000+ authorized AI devices are not generative AI and thus represent a different regulatory challenge is validated by the provided sources and aligns with pub

Sources cited: 11. Validation challenges: 0. Review cost: about $0.04. Learn how these briefings are written and verified.