Artificial intelligence is no longer a pilot project or future investment. It is actively shaping clinical decision-making and is increasingly embedded in the medical devices that clinicians rely on every day. The majority of these devices are concentrated in radiology and image‑analysis applications, followed by cardiology, neurology, and other diagnostic specialties, according to the U.S. Food and Drug Administration (FDA). From radiology workflows to surgical navigation systems, AI-enabled tools are influencing diagnoses, guiding procedures, and, in some cases, determining the trajectory of patient care in real time. For healthcare leaders focused on advancing value-based care, this shift presents both a strategic opportunity and a growing source of clinical and enterprise risk.

Rapid Growth of AI in Medical Devices and Marketing Authorization

The scale of AI adoption is striking. In 2015, the FDA had authorized only a small number of AI-enabled medical devices. By the end of 2025, that number had surpassed 1,400, with nearly 300 devices authorized by the FDA in a single year. Most of these devices were authorized through the 510(k) clearance pathway, which enables faster market entry by demonstrating substantial equivalence to existing technologies. Adoption has been concentrated in medical imaging, where three-quarters of AI-enabled devices are currently used, but use in procedural settings and real-time clinical decision support is rapidly expanding.

For health systems, this rapid growth is occurring alongside broader digital transformation efforts. AI is being layered into enterprise strategies that include predictive analytics, virtual care, and clinical workflow optimization. For example, a health system might use AI to flag patients at elevated risk of deterioration, route those patients into virtual monitoring programs, and surface real‑time recommendations within the clinician’s existing workflow. Unlike traditional health IT tools, however, AI-enabled medical devices operate directly within clinical decision-making. This distinction elevates both the potential positive impact and the associated risks of these products.

A central challenge of the AI-enabled medical devices surge is the gap between regulatory clearance and real-world performance. FDA clearance of a 510(k) reflects a determination that a device is “substantially equivalent” to a legally marketed device (i.e., it is as safe and as effective as another device that has marketing authorization). It is not an independent, stand‑alone determination by the FDA that the device is safe and effective on its own merits, and it does not guarantee consistent performance across diverse clinical environments. AI models are particularly sensitive to variations in data, workflow, and patient populations. Health systems that assume uniform performance may encounter unexpected variability in outcomes.

Adverse Outcomes, Patient Injuries, and Emerging Litigation

Recent reports highlight the consequences of the emerging gap between expectations and outcomes. One widely discussed example involves the TruDi Navigation System, an AI-enhanced surgical navigation device used in sinus and skull-base procedures. Following the integration of machine‑learning functionality into the system’s software, the FDA’s post‑market surveillance data reflected a marked increase in reported malfunctions and adverse events. Reported complications included cerebrospinal fluid leaks, vascular injuries, and strokes, often associated with inaccurate instrument localization during procedures. More broadly, post‑market analyses have identified a rising number of AI‑enabled medical devices linked to product recalls, many occurring within the first year following authorization. Together, these developments underscore the limitations of premarket review alone and highlight the need for robust post‑deployment validation, monitoring, and governance at the health‑system level when AI functionality is incorporated into clinical technologies.

The implications of AI liability exposure extend beyond clinical performance and capture a broader enterprise risk. As AI becomes more deeply integrated into care delivery, health systems will have to assume a more active role in the lifecycle management of these technologies. Liability is no longer confined to manufacturers — providers and health systems will face heightened exposure and scrutiny related to implementation decisions, clinician training, oversight failures, and informed consent practices. A body of case law involving professional negligence and vicarious liability has already begun to take shape in response to these developments.

Courts and regulators are beginning to grapple with these issues in determining how much risk patients can reasonably be expected to assume and how much must be mitigated through design, oversight, and disclosure. Traditional liability frameworks have historically focused on product defects, and neither those frameworks nor traditional medical malpractice doctrines were developed with adaptive, probabilistic software systems in mind. As a result, courts face increasing difficulty determining whether liability should rest with device manufacturers, clinicians, healthcare institutions, or some combination thereof. These challenges are compounded where traditional theories of liability are supplemented by claims alleging inadequate validation, insufficient disclosure, or overreliance on algorithmic outputs. At the same time, federal regulators have signaled increased attention to post-market performance, transparency, and lifecycle oversight for AI-enabled devices.

Recent FDA guidance on clinical decision support software, finalized in early 2026, reinforces that not all AI tools will be subject to active regulatory oversight, particularly those intended to support rather than replace clinician judgment. This distinction places greater responsibility on health systems to evaluate performance, ensure appropriate use, and manage risk for tools that may fall outside traditional regulatory controls.

Heightened Safety Protocols, Assumption of Risk, and Informed Consent

For organizations advancing value-based care strategies, this creates a critical inflection point. While AI has the potential to improve key performance metrics, such as diagnostic accuracy, length of stay, readmission rates, and cost per patient episode, those benefits are not guaranteed. Without appropriate safeguards, AI can introduce new sources of variability that may undermine performance and increase downstream costs.

A disciplined and structured approach to AI governance is essential. Leading organizations are beginning to treat AI-enabled devices not simply as technology acquisitions, but as clinical interventions that require ongoing oversight. This includes establishing multidisciplinary governance structures with thorough policies that bring together clinical leadership, data science, compliance, information technology, and legal advice.

Continuous performance monitoring is emerging as a foundational capability. Health systems are exploring how well AI tools perform across different patient populations and care settings, using real-world data to identify drift, bias, or degradation in performance. Evidence shows that AI models may experience measurable declines in accuracy when applied outside their original training environments, reinforcing the importance of local validation prior to widespread deployment and ongoing scrutiny of AI Oversight Committees to ensure consistent long-term results.

Equally central to the impact of AI on health systems is the role of clinicians. AI is most effective when it augments, rather than replaces, clinical judgment. Yet the risk of a phenomenon called automation bias (the inclination of humans to favor decisions generated by AI systems) presents a well-documented risk to clinician judgment and patient well-being. To mitigate that risk, health systems must ensure that AI tools are implemented in a way that supports informed decision-making, including transparent communication of a tool’s confidence levels and limitations.

Patient engagement also warrants greater attention. Several lawsuits and investigative reports have noted that patients were allegedly unaware that AI‑enabled systems would be used in their care or that such systems carried distinct risks. As transparency and consent become exponentially more important components of trust and risk management, health systems should explore more explicit informed consent processes to explain algorithmic uncertainty, data limitations, and the potential for error in machine‑generated outputs.

Conclusion

From a strategic perspective, the integration of AI-enabled medical devices should be closely aligned with value-based care objectives, as the industry continues its transition from volume to value. Health systems should assess whether AI tools contribute to measurable improvements in outcomes, reductions in unnecessary utilization, and overall cost efficiency.

While these devices represent a significant advancement in the ability to deliver more precise, data-driven care, they also introduce new complexities that require equally sophisticated approaches to governance, oversight, and clinical integration. For healthcare executives, the mandate is clear: AI must be managed with the same rigor applied to any clinical intervention. Organizations that succeed in doing so will be better positioned to realize the promise of AI while safeguarding patient outcomes and maintaining trust.