FDA Considers Letting Developers Update Radiology AI Without New Reviews

The FDA issued a Federal Register notice on December 29, 2025. It requests public comments through February 27, 2026, on a citizen petition. The petition seeks partial premarket notification exemptions for some class II radiology AI software. Rubrum Advising submitted the petition in October 2025 on behalf of Harrison.ai (Docket FDA-2025-P-5560). It targets radiological computer-aided detection and diagnosis product codes POK, MYN, QAS, QFM, QBS, and QDQ. The petition asks to allow streamlined updates after an initial 510(k) clearance. Vendors must maintain prior regulatory compliance, robust post-market monitoring, change transparency, and clinician training. Quality systems, special controls, and registration and listing would remain required. Radiology accounts for over 75% of the FDA's 1,357 authorized AI-enabled devices. The proposal would let models evolve without full resubmissions. It would also improve real-world surveillance for drift, false positives, and patient outcomes. Separate January 2026 FDA guidances loosen oversight for low-risk wellness wearables and clinical decision support.

Why It Matters

• ML engineers at medtech firms can iterate cleared SaMD models faster using streamlined updates instead of repeated 510(k) submissions, if prior clearance and approved post-market plans are in place.
• Healthcare IT teams should implement continuous performance monitoring and cybersecurity controls for production SaMD to detect model drift and vulnerabilities from deployment.
• Radiology PACS administrators can expect faster AI tool updates but should log clinician overrides and outcome metrics to support FDA audits and post-market surveillance.
• Compliance teams should submit comments by Feb 27, 2026, to influence potential partial-exemption criteria and rollout timelines in 2026.