The US Food and Drug Administration (FDA) has prioritised guidance on AI-enabled medical devices, generative AI in mental healthcare and robotic surgical systems for fiscal year 2027. Key areas include AI lifecycle management, software updates and premarket evidence for surgical robots, alongside recommendations for generative AI conversational devices intended for mental disorders.
The priorities come as the FDA’s public list recorded more than 1,600 AI-enabled medical devices authorised for marketing by September 2026, spanning imaging, cardiology, neurology and surgical applications. Companies in this space include GE HealthCare, Philips, Siemens Healthineers and AI developers Aidoc and Annalise.ai.
As AI systems become more complex and medical robots integrate software, instruments and human control, regulators face a growing challenge: how to enable technological advances while maintaining safety, effectiveness and regulatory oversight.
AI Medical Devices: Why the Regulatory Approach Is Evolving
The FDA’s AI-enabled medical device list illustrates how widely machine learning has entered medical technology. It includes diagnostic imaging software, image reconstruction tools, cardiac assessment systems, retinal image analysis and technologies that support treatment decisions.
These applications are no longer limited to standalone software developed by specialist companies. AI is increasingly embedded in medical equipment supplied by established manufacturers and used by hospitals, diagnostic centres and other healthcare providers.
The figure of 1,600-plus devices requires context. It represents authorised devices, not 1,600 independent companies, and does not mean that every device uses generative AI. The list is updated periodically, so its total depends on the records available at the time of reporting.
The FDA generally regulates these products according to their intended use, technological characteristics and risks. Depending on the device, manufacturers may use one of three major premarket pathways:
- 510(k) premarket notification: Generally used to demonstrate that a device is substantially equivalent to a legally marketed predicate device.
- De Novo classification: Provides a route for certain novel, low-to-moderate-risk devices for which there is no suitable predicate.
- Premarket approval (PMA): The more demanding pathway for devices requiring that level of review, including applicable high-risk devices.
AI does not automatically create a separate regulatory pathway. The applicable route depends on the device and its intended use, including the risks associated with its AI functionality.
As the technology develops, manufacturers must establish not only that their products work as intended but also that their testing, documentation and risk controls are appropriate for the proposed clinical application.
AI Lifecycle Management: Why Initial Validation Is Not Enough
The FDA plans to finalise its draft guidance, Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations, published in January 2025. The guidance addresses lifecycle documentation and risk management, including data quality, performance evaluation, transparency, bias, monitoring and maintenance.
For manufacturers, the regulatory implications extend to the evidence submitted for marketing authorisation. Validation data should support the device’s intended use, target population and operating conditions. Manufacturers also need to document model performance, evaluate relevant sources of bias, identify limitations and explain how these factors affect clinical use.
Post-market monitoring is another key consideration. Manufacturers should establish processes to identify performance changes, assess complaints and investigate potential safety concerns. If a model update could affect safety or effectiveness, the manufacturer must determine whether the change falls within an authorised or established predetermined change control plan (PCCP), where applicable, or requires a new FDA marketing submission. The regulatory pathway depends on the nature and impact of the change.
Finalising the guidance could provide manufacturers with clearer recommendations for preparing marketing submissions and documenting lifecycle risk management. It may also influence how companies plan validation studies, manage software updates and maintain evidence supporting device performance after market entry. The guidance itself would set out FDA recommendations; applicable statutory and regulatory requirements would continue to govern manufacturers’ obligations.
Software Updates: The Role of Predetermined Change Control Plans
AI systems may require modifications after authorisation. Manufacturers could need to correct performance problems, improve an algorithm or introduce changes anticipated during development.
The FDA’s final guidance on predetermined change control plans (PCCPs) for AI-enabled device software functions, published in August 2025, provides an existing framework for certain planned modifications. The agency’s 2027 agenda also prioritises finalising broader guidance on PCCPs for medical devices.
A PCCP describes anticipated modifications, the methods used to develop and validate them, and the procedures for implementing them. When included in an appropriate authorised marketing submission, a PCCP can allow specified changes to be implemented without a separate marketing submission for every change covered by the plan.
For example, a manufacturer developing an AI-enabled imaging device might anticipate a defined performance improvement. Its plan could specify the proposed modification, testing procedures, acceptance criteria and controls required before implementation.
The benefit is greater predictability. Manufacturers can plan for certain changes in advance rather than assessing every anticipated modification as an entirely new regulatory exercise.
However, a PCCP is not unrestricted permission to retrain or alter a model. The changes must remain within the authorised plan and satisfy its conditions. Modifications outside its scope may require additional regulatory assessment.
This distinction is particularly important for products marketed with promises of continual improvement. A company’s ability to update its software technically does not establish that every update is acceptable from a regulatory perspective.
For developers, PCCPs connect engineering decisions directly with regulatory strategy. The anticipated changes, validation methodology, risk assessment and implementation controls must be considered together.
Generative AI in Mental Healthcare: What Evidence Will Be Needed?
The FDA plans to develop draft evidentiary recommendations for generative AI conversational devices intended to treat mental disorders. These products raise distinct safety concerns because responses can vary and inappropriate outputs may affect patient decisions or delay professional care.
An August 2026 FDA discussion paper examines risk assessment, premarket evaluation and post-market monitoring, including challenges posed by foundation models and agentic AI systems.
The paper explores competency-based evaluation, potentially combining non-clinical benchmarking with clinical confirmation. This could help establish whether a system is suitable for its intended clinical use, beyond simply generating fluent responses.
Regulatory status depends on a product’s intended use and functionality. A general-purpose chatbot does not automatically become a medical device because it is used to discuss mental health.
The discussion paper is not draft or final guidance and does not establish new regulatory expectations. It seeks feedback to inform the FDA’s approach. Future recommendations may clarify how developers should demonstrate response reliability, assess harmful outputs and substantiate clinical suitability.
Surgical Robots: Testing, Clinical Evidence and Labelling
The FDA’s surgical-robot priority is supported by a separate draft guidance on Robotically-Assisted Surgical Devices: Premarket Submissions.
The September 2026 draft addresses non-clinical testing, clinical data and labelling for certain robotically assisted surgical devices. These systems use software-controlled robotic components to help qualified practitioners position and control surgical instruments during procedures.
Their safety assessment must consider the interaction between mechanical components, software, instruments, control interfaces and the surgeon. Testing only one component may not adequately demonstrate the safety and effectiveness of the complete system.
Non-clinical testing can evaluate mechanical performance, system reliability and software behaviour. Clinical data, where appropriate, can support conclusions about safety and effectiveness for the intended use. Labelling must communicate indications, limitations and information necessary for safe operation.
For example, manufacturers need to consider how a robotic platform responds to faults, how the operator maintains control and whether the available safeguards are appropriate for foreseeable risks.
The regulatory questions become more complex as systems incorporate additional automation or remote-control capabilities. A system that assists a surgeon with instrument positioning raises different issues from one that performs a task with limited direct intervention.
The FDA is also examining broader benefit-risk considerations for robotic medical devices with autonomous or remote teleoperation capabilities. Its announced public workshop on these topics is scheduled for 2–3 December 2026, as a separate initiative from the surgical-robot premarket draft guidance.
Manufacturers should review the draft guidance against their existing testing and submission plans. The comment deadline is 24 November 2026, and the document remains non-binding while in draft form.
What the FDA’s Guidance Programme Numbers Reveal
The 2027 agenda should be read alongside the agency’s record of publishing guidance. The FDA’s Center for Devices and Radiological Health reported that it published 7 of the 11 documents on its fiscal year 2026 guidance-priority lists: five from the A-list and two from the B-list.
The A-list identifies guidance the agency intends to publish during the fiscal year. The B-list identifies guidance it plans to publish as resources permit. An additional “Under Construction” list covers documents the agency intends to develop as resources allow.
These categories matter for companies planning regulatory submissions. A topic on a priority list signals agency attention, but it does not guarantee a particular publication date or a specific final recommendation.
The FDA has also committed, as resources permit, to take action on 80% of draft guidance documents within three years of the close of their comment periods and 100% within five years. “Action” can include finalising or withdrawing a document, reopening the comment period or issuing a new draft. It does not necessarily mean that every draft will become final guidance.
The agency is accepting feedback on its fiscal year 2027 guidance-priority lists until 30 November 2026. This gives manufacturers, clinicians, researchers and trade associations an opportunity to identify evidence gaps, explain implementation challenges and recommend changes to the planned programme.
What Medical Device Companies Should Do Now
The agenda has practical implications for manufacturers at different stages of product development.
- AI developers should review validation and monitoring. Companies should assess whether their testing covers the intended patient population and use environment, and whether they have processes to detect unexpected performance changes after deployment.
- Companies planning AI updates should examine their change-control strategy. A PCCP may provide a structured route for anticipated modifications, but its scope, validation methods and implementation controls must be clearly defined.
- Generative AI developers should substantiate clinical claims. Companies need to distinguish general conversational functionality from a medical purpose and assess the risks associated with their intended use. They should monitor the FDA’s work on evidentiary recommendations rather than assume that the discussion paper establishes new requirements.
- Surgical-robot manufacturers should review their complete evidence package. Mechanical testing, software validation, clinical evidence and labelling must support the safety and effectiveness of the system for its proposed use.
Companies should also consider early engagement with the FDA where the regulatory pathway or evidence strategy is uncertain. The appropriate pathway and submission requirements will depend on the specific device, its intended use and its risk profile.
Outlook
The FDA’s 2027 guidance agenda reflects a shift in regulatory attention from evaluating individual technologies at a single point in time to considering how complex medical systems are developed, changed and monitored throughout their lifecycle.
The scale of adoption, with more than 1,600 AI-enabled medical devices authorised for marketing, underscores the importance of consistent approaches to evidence, software changes and post-market monitoring. The agency’s priorities also show that generative AI and robotic surgery raise distinct questions that cannot be addressed by a single, universal framework.
For manufacturers, the immediate priority is to build robust evidence strategies around existing requirements, prepare for controlled software changes and follow the development of relevant guidance. The final documents may clarify regulatory recommendations, but their practical impact will depend on their final scope and the characteristics of each device. Companies that distinguish planned guidance from binding requirements will be better placed to manage submissions, product updates and regulatory risk.
SOURCES:
- https://www.fda.gov/medical-devices/guidance-documents-medical-devices-and-radiation-emitting-products/cdrh-proposed-guidances-fiscal-year-2027-fy-2027
- https://www.fda.gov/regulatory-information/search-fda-guidance-documents/artificial-intelligence-enabled-device-software-functions-lifecycle-management-and-marketing
- https://www.fda.gov/medical-devices/digital-health-center-excellence/artificial-intelligence-enabled-medical-devices
- https://www.fda.gov/regulatory-information/search-fda-guidance-documents/marketing-submission-recommendations-predetermined-change-control-plan-artificial-intelligence
- https://www.fda.gov/medical-devices/software-medical-device-samd/predetermined-change-control-plans-machine-learning-enabled-medical-devices-guiding-principles
- https://www.fda.gov/news-events/press-announcements/fda-issues-comprehensive-draft-guidance-developers-artificial-intelligence-enabled-medical-devices
