Trends in AI/ML-Enabled Device Authorizations by FDA: A Shift from Growth to Maturity

By Shannon Campbell, PhD, Principal Consultant, Frank Healthcare Advisors

FDA recently updated its Artificial Intelligence-Enabled Medical Devices List (on April 2, 2026), providing insights into regulatory pathways and real-world examples of successful submissions. FHA’s analysis indicates that FDA’s regulatory activity around AI/ML-enabled medical devices is no longer just accelerating, it is maturing, diversifying, and becoming more predictable. For companies developing AI-enabled devices and diagnostics, this shift is critical. The question is no longer “Can AI get through FDA?” but rather “What does success now look like, and how do we build for it?”

1) FDA Authorization Volume: From Acceleration to Scale

The FDA’s AI-enabled device ecosystem continues to expand at a remarkable pace:

• Over 1,350 AI-enabled devices are now authorized for marketing in the U.S.

• Annual clearances have climbed steadily, with ~300 new authorizations in 2025 alone 

• The 510(k) pathway still dominates (>95%), reinforcing the importance of predicate-based strategies

This confirms what many in the industry are experiencing firsthand:AI is no longer experimental, it is operational. But more importantly, the consistency in clearance volume suggests something deeper:

• FDA expectations for AI/ML devices are becoming increasingly well understood and reproducible.

2) Modality Concentration Persists, But Use Cases Are Expanding

Radiology continues to dominate AI device authorizations, accounting for the majority of cleared products. However, recent FDA activity signals gradual diversification:

• Growth in cardiology, ophthalmology, and neurology applications

• Expansion into workflow tools, triage systems, and quantitative analysis platforms

• Increasing presence of non-image-based AI, including diagnostics and clinical decision support

This trend reflects a broader shift:

• AI is moving from image interpretation to clinical workflow integration and decision support

For innovators, this creates both opportunity and complexity, particularly as clinical validation expectations vary significantly by use case.

3) PCCPs: The Beginning of True Lifecycle Regulation for AI

One of the most important regulatory developments in the past year is the operationalization of Predetermined Change Control Plans (PCCPs).

Following FDA’s final guidance (August 2025), PCCPs are now appearing in cleared devices:

• Example: Fibresolve (K252041) incorporated a PCCP to allow pre-authorized algorithm updates

• Additional late-2025 clearances (e.g., radiology platforms) demonstrate similar adoption patterns

This is a pivotal shift. Historically, AI models were effectively “locked” at clearance. With PCCPs:

• Manufacturers can pre-specify future modifications

• FDA evaluates the process of change, not just the static model

• Updates can occur without new submissions, if within defined bounds

This marks FDA’s transition toward a Total Product Lifecycle (TPLC) regulatory model for AI. For companies, this creates a new strategic requirement:

• You are no longer just submitting a device, you are submitting a controlled evolution pathway.

4) Real-World Evidence (RWE): From Supplementary to Strategic

Alongside PCCPs, Real-World Evidence (RWE) is becoming increasingly central:

• FDA’s updated December 2025 RWE guidance supports use of aggregate data for regulatory decisions

• RWE has now contributed to hundreds of device authorizations, including AI-enabled systems

• In AI contexts, RWE is being used for:

  • Post-market performance monitoring

  • Drift detection and model stability

  • Supporting PCCP validation frameworks

This reflects a fundamental shift in evidence strategy: Clinical validation is no longer confined to premarket studies, it is continuous and lifecycle-based.

For AI developers, this reinforces the importance of:

• Data infrastructure

• Performance monitoring systems

• Real-world deployment strategies

5) De Novo and Novel AI: Still the Exception, Not the Rule

Despite rapid growth, one trend remains unchanged, truly novel AI devices still face higher regulatory friction:

• The 510(k) pathway dominates, reinforcing incremental innovation

• De Novo pathways remain limited, particularly for:

  • Foundation model–based systems

  • Autonomous decision-making tools

  • Novel, non-imaging AI applications

This creates a structural tension between the regulatory system that favors predicate-based evolution and the technology that is moving toward foundational, adaptive, and generalizable AI.

Until further guidance emerges, companies pursuing novel AI architectures should expect:

• Higher evidentiary burden

• Greater reliance on FDA engagement (e.g., Pre-Subs, TAP)

• Increased importance of regulatory strategy early in development

6) What This Means for AI-SaMD Developers

The current FDA landscape sends a clear message that success is no longer about novelty, it’s about execution. Companies that succeed are those that:

• Align early with established regulatory patterns (e.g., 510(k), assistive AI positioning)

• Design for lifecycle management (PCCPs, post-market monitoring)

• Build evidence strategies that extend beyond clearance

At the same time, emerging expectations are raising the bar:

• Transparency around AI behavior and limitations

• Bias mitigation and dataset representativeness

• Robust software lifecycle and risk management under QMSR

7) Strategic Implications for the Ecosystem

For Startups

• Prioritize regulatory pathway clarity early, this is now a gating factor for investment

• Consider PCCP strategy upfront, not as an afterthought

• Invest in real-world data infrastructure from day one

For Investors

• Evaluate companies on regulatory readiness and lifecycle strategy, not just technical innovation

• Look for teams that understand FDA expectations for AI explainability, validation, and monitoring

For Regulators

• Continue expanding clarity around:

  • Adaptive AI and foundation models

  • Use of RWE in premarket submissions

  • Scalable frameworks beyond predicate-based pathways

Final Thought

The FDA’s approach to AI/ML-enabled devices is no longer in its infancy, it is entering a phase of structured maturity where:

• Authorization volumes are high

• Regulatory expectations are stabilizing

• Lifecycle oversight is becoming the norm

But the next challenge is already emerging:

• How do we regulate AI that is not just adaptive, but continuously learning, generalizable, and deeply integrated into clinical decision-making?

The companies that succeed in this next phase will not just build better algorithms, they will build regulatory-grade AI systems designed for trust, transparency, and evolution.