AI-enabled medical devices: what FDA's list and guidance show

How many AI-enabled medical devices has FDA authorized?

FDA publishes a list of AI-enabled medical devices that have been authorized for marketing in the United States and updates it periodically. An analysis of the list published in September 2026 counted 1,614 devices through the end of June 2026, up 5.9% on the first quarter. FDA authorized 92 AI-enabled devices in the first quarter of 2026 and 89 in the second.

FDA describes the list as a resource for the public and not a comprehensive record. Devices are identified from AI-related terms in the public authorization summaries, so a device that uses AI without saying so in its summary may be missing.

Which specialties do they serve?

Radiology dominates. Of the 1,614 devices, 1,230 are radiology devices, which is 76%. Image analysis suits machine learning: the input is digital, large labelled datasets exist, and the output can be checked against a reader. Cardiovascular and neurology devices follow at a distance.

What is a predetermined change control plan?

A predetermined change control plan, or PCCP, is a part of a marketing submission in which the manufacturer describes the modifications it intends to make to an AI-enabled device software function after authorization, and how it will make them. Congress gave FDA the authority to accept such plans in 2022, in section 515C of the Federal Food, Drug, and Cosmetic Act. FDA's guidance on PCCPs for AI-enabled device software functions is final; it was first issued in December 2024 and the current version is dated August 2025.

Once FDA has authorized a device with a PCCP, the manufacturer can implement the changes the plan covers without a new submission for each one, provided it follows the plan.

What does the draft lifecycle guidance ask for?

In January 2025 FDA published the draft guidance Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations. It takes a total product life cycle view and lists what a marketing submission should contain.

• A device description that says what the AI does, its inputs and outputs, and who uses it.
• User interface and labelling that tell the user what the model was trained on, how it performs and where its limits are.
• A risk assessment that covers AI-specific risks, including bias and performance that varies between patient groups.
• Data management: how training and test data were collected, kept separate and shown to represent the intended population.
• Model description, development and validation, with performance broken down by relevant subgroups.
• A plan for monitoring performance after the device is on the market.

Where do robotics fit in?

Robotically-assisted surgical devices are regulated as medical devices in their own right, and a growing number add AI features such as image guidance and motion planning. When they do, the same expectations apply to the AI function: a defined intended use, evidence of performance, transparency to the user and a plan for changes.

On the manufacturing side, device makers now work under the Quality Management System Regulation, which took effect on 2 February 2026 and incorporates ISO 13485:2016. Software used in production and in the quality system, AI-enabled or not, is validated for its intended use with effort proportionate to risk.

What should a device quality team do now?

• Decide early whether the model will change after launch. If it will, plan the PCCP with the first submission instead of retrofitting it.
• Treat training and test data as controlled records, with provenance, separation and representativeness documented.
• Build post-market performance monitoring into the quality system, with complaint handling and CAPA connected to it.
• Apply the ten Good Machine Learning Practice guiding principles that FDA, Health Canada and the MHRA published in 2021 as a design checklist.