How AI-enabled Software as Medical Devices are Reshaping Regulatory Routes and Medicine

How AI-enabled Software as Medical Devices are Reshaping Regulatory Routes and Medicine August 17, 2026 | 6 min read | JoAnne Bronikowski, Devin Ridgley Twitter LinkedIn Facebook Email Print As AI-enabled Software as Medical Devices become more prevalent, regulatory agencies are responding with new guidelines and AI-specific authorisation routes. This article looks at current trends, suggests clinical considerations for medical device developers and anticipates AI-enabled SaMDs’ future progress. FDA authorisations of AI-enabled software as medical devices (SaMDs) have increased significantly in recent years.

Over 1,500 AI-enabled medical devices were FDA authorised as of April 2026. The majority (97%) of those authorisations happened within the preceding 10 years, with 68% authorised since 2022. Radiology has dominated, accounting for 76% of authorisations, followed by cardiovascular (10%) and neurology (4%).

With the expansion of AI’s capabilities, regulatory agencies are adapting leading to a more complex, demanding environment for medical device developers. Previously regulators authorised AI-enabled SaMDs under De Novo classification as the SaMDs were low-to-moderate risk novel devices with no legally marketed predicate devices that could support a 510(k) substantial-equivalence claim. However, as these De Novo classifications establish new device classifications, regulators are increasingly able to use the 510(k) pathway by claiming equivalence to these newly established predicates.In addition, regulators are modernising their frameworks to account for the unique profile and potential impacts of this revolutionary technology.

Here, we explore recent trends, global regulatory considerations, offering guidance on FDA Pre-Sub meetings and clinical considerations for AI in SaMD developers. Recent trends in AI-enabled authorisations Current AI-enabled medical devices are progressing from detection, to triage, to diagnosis, and ultimately toward treatment guidance. These trends indicate that AI systems are advancing beyond simply identifying conditions to shaping individual treatment strategies.

We see the same evolution across technologies regardless of the application: wearables, IVDs and radiology all follow this trajectory.Wearables have advanced rapidly with AI-enabled technologies progressing from the early vital sign monitoring of 2019’s Pera Watch to clinical-grade AFib-specific wrist devices and, most recently, large scale consumer smartwatch AFib screening features in Apple, Samsung and Fitbit. AI-enhanced patch-based wearable devices have also evolved, the AT-patch system couples a torso worn ECG recording patch with AI-based analysis software. Long term ECG recordings are processed to detect arrythmias and produce a physician-reviewed diagnostic support for adults with transient cardiovascular symptoms including palpitations, dizziness or shortness of breath.While there has been a limited number of FDA clearances of IVD AI tools, they are emerging in genomic profiling, host response assays and infectious disease triage.

One of the most recent FDA authorisations exemplifies AI’s power when coupled with state of the art assays. In 2025 the FDA authorised the TriVerity test, an AI-enabled IVD that analyses patient whole blood samples of RNA genes associated with the host-response to infections. The intended use is to aid triaging suspected sepsis in emergency department patients.

The AI-algorithm is designed to differentiate between bacterial, viral or non-infectious agents based on the RNA profile of the patient’s response genes to the infection.We anticipate that AI-enabled medical technologies will develop to become integrated into decision-making based on multimodal data. Cancer care is the most likely area to first adopt this approach. It is not science fiction to foresee the combination of mammography results with biopsy-derived genomic data used to characterize a tumour and recommend an individualised treatment.

Reaching this level of maturity will require clinical trials that demonstrate that AI-driven diagnostic and treatment decisions are equivalent or superior to physician-directed care. Global regulatory considerations Regulatory agencies worldwide continue to tighten requirements for AI-enabled SaMD, adapting existing medical device frameworks to address higher levels of autonomy, clinical influence and lifecycle change. Developers need to plan for divergence across major markets while building common evidence foundations.

In the EU, MDR/IVDR requirements sit alongside the EU AI Act, which classifies many AI-enabled medical devices as high risk. The UK MHRA, Health Canada, Japan’s PMDA and MHLW, China’s NMPA, Singapore’s HSA and Australia’s TGA all apply risk-based approaches similar to the US FDA. These regulatory authorities vary in how they manage adaptive algorithms, change control and post-market performance.

We recommend defining intended use and clinical risk early. Developers should clarify predicate rationale, build representative datasets, and document algorithm architecture and performance criteria. As with any new treatments, regulators expect robust quality systems, cybersecurity controls, real-world performance monitoring along with evidence that AI outputs support safe, clinically meaningful decisions.

FDA Pre-Sub guidance The Pre-Submission process is the first major regulatory touchpoint in the US. For medical device developers it is an essential step to reduce regulatory uncertainty and ensure that development activities are positioned for a successful submission. Pre-Sub questions provide valuable insights into a range of topics including device software function classification, device validation parameters, algorithm design and more.

Typical topics include indications for device use/intended use, predetermined change control plan (PCCP) and human factors of the study design. Pre-Sub meetings are limited to an hour so developers should plan to maximise this interaction. Common pitfalls are broad, vague or theoretical questions, or submissions that address too many topics to cover within the meeting.

We advise teams to plan their questions strategically, appoint meeting roles and assign a note-taker. Team members should be familiar with current FDA guidances and respond promptly to FDA requests for information. Clinical considerations for AI in SaMD design It is widely acknowledged that AI models can have biases.

Depending on the intended use developers may need to avoid or exploit them. For example, an intentional pediatric patient bias is appropriate for an SaMD designed to treat pediatric patients. However, unintentional biases in medical device datasets should be prioritised for reduction or elimination.

Developers must ensure that training, validation and testing datasets are representative of the target patient population. Lack of or misproportioned data inclusion in the AI training datasets can result in the omission of underrepresented patient groups from the AI model. The presence of inaccurate labels can also introduce unintended bias or significantly reduce the model’s accuracy.

ICON’s experience In the last seven years ICON has supported 17 trials for digital therapeutics and AI-enabled SaMDs across a range of areas including cardiovascular, psychiatry, vision, pulmonary, and diagnostic imaging and testing. These trials involved more than 1,700 participants and healthy volunteers in over 100 sites in North America and Europe. ICON has the expertise, agility and collaborative skills that are essential for medical device and diagnostics development.

We offer comprehensive regulatory services, global clinical study design and execution by dedicated medical device experts. Partner with ICON to bring your product to market faster.This article draws from our series of whitepapers for developers of AI-enabled SaMDs, find them here.Sponsored content is written and provided to BioSpace by the advertiser. It is published with the advertiser’s approval without contribution from BioSpace’s editorial and insights teams.

Twitter LinkedIn Facebook Email Print Sponsored JoAnne Bronikowski Senior Manager, Regulatory AffairsRegulatory Affairs Medical Devices, ICON Devin Ridgley Director of Project ManagementMedical Device & Diagnostics Research, ICON