What 36 Authorized PCCPs Reveal About AI Device Monitoring in 2026
By Sentir Health · 06/23/2026
We read every FDA-authorized Predetermined Change Control Plan cleared through the 510(k) pathway so far in 2026, going through the clearance letters and summaries one at a time. The headline count is climbing fast, but it tells you the least about what is actually going on.
The pace is real
Between January 9 and May 27, the FDA authorized 36 PCCPs through the 510(k) pathway. For comparison, 104 had been authorized in total across every year before 2026, and only 54 of those had come by the middle of 2025. So in under five months, with most of the year still ahead, 2026 has already produced more than a third of the entire prior history of the mechanism. The real figure is higher still, because the FDA posts these documents on a delay and the most recent months are incomplete.
Most people in the field can feel that acceleration. The more useful question is what those 36 devices actually are.
Most PCCPs are not AI software
Sort the 36 by what the device really is and they fall into three roughly equal groups.
AI/ML software-as-a-medical-device accounts for 12. These are the standalone algorithms, the devices that carry the ongoing performance-monitoring obligation the framework was built for.
AI-enabled hardware and systems account for another 12. Here the intelligence lives inside a physical product: electrophysiology navigation and catheter-mapping systems, insulin pumps with predictive control, robotic surgical platforms. The PCCP often covers device and component changes as much as it covers a learning model.
Non-AI devices make up the last 12. These are traditional hardware and component changes, plus deterministic, rules-based software like a glycemic-management calculator or antimicrobial-susceptibility test panels. There is no model to drift and nothing to monitor in the machine-learning sense.
So only a third of 2026's PCCPs are what most people picture when they hear "AI medical device."
AI software is not what is driving the growth
Compare the same January to May window in each year, matched five-month periods rather than a full year against a half.
In 2025 the FDA authorized 29 PCCPs in that window: 13 AI/SaMD, 1 AI-enabled hardware, and 15 non-AI. In 2026 it authorized 36: 12 AI/SaMD, 12 AI-enabled hardware, and 12 non-AI. AI/SaMD software stayed flat, going from 13 to 12. Non-AI devices fell, from 15 to 12. Almost all of the growth came from AI-enabled hardware, which went from a single PCCP in that window a year ago to twelve this year.
So the thing pushing PCCP numbers up in 2026 is not AI software, and it is not traditional devices either. It is hardware with AI inside it: insulin pumps with predictive control, navigation and mapping systems, surgical robotics. And since the most recent months are still being posted, the 2026 counts will only rise from here.
Two kinds of company in the AI group
The 12 AI/SaMD plans come from 11 companies, with Anumana holding two of this year's twelve. Those companies sit in two very different positions.
A few are large strategics, or the AI units of larger companies. They have regulatory departments, in-house machine-learning infrastructure, and the resources to build whatever monitoring they decide they need. The rest are independent AI companies, several of them building an entire business on a portfolio of algorithms that keep improving. For those companies, monitoring real-world performance is not a project that ends. It is a standing operating cost that grows with every device and every version, and there is no large compliance team to absorb it.
A plan is not the system that keeps it
The most important finding is also the most concrete. Of the 64 AI/SaMD PCCPs in our tracker, only 8 include an explicit real-world performance-monitoring commitment in the cleared documentation. Most state none at all, and a handful are statement filings whose contents are not public.
That gap is the whole story. A PCCP records what a company promised the FDA. It says nothing about whether the company has built the system that can deliver on it: the kind that runs continuously, catches the failures that do not announce themselves, and holds up when an auditor or a hospital asks how you know the model is still performing as cleared.
We made the fuller version of that case in The PCCP Inflection Point. The short version is that the plans are multiplying, and for most companies the monitoring is still being figured out.
Find your device in the data
The Sentir PCCP Tracker lists every FDA-authorized PCCP and lets you browse by category, clinical panel, PCCP type, or company. If you hold one, you can see where it sits in the landscape and what its monitoring obligation implies.
Sentir Health is the independent system of record for the performance of FDA-cleared AI. We produce the audit-defensible evidence your PCCP requires. Learn more or book a call.
Methodology note: Counts come from FDA 510(k) and De Novo records via the Sentir PCCP Tracker. Figures are current as of 2026-06-17 and cover the 510(k) pathway except where noted. Recent authorizations may lag until the FDA posts their documents. PCCP type, category, and monitoring-commitment labels are editorial classifications based on public documents, not official FDA designations.