Artificial intelligence in Europe's medical device database: Public traceability of AI-enabled software in EUDAMED.

BACKGROUND: AI-enabled medical software is increasingly used across clinical care. The European Database on Medical Devices (EUDAMED) is intended to support transparency, but the public traceability of AI functionality remains uncertain. OBJECTIVE: To assess AI traceability in EUDAMED software-device records and to characterize devices supported by explicit device-specific AI evidence. METHODS: We conducted a cross-sectional study of EUDAMED software records available through 31 December 2025. Consolidation yielded 1546 device-level records from 2858 raw records. EUDAMED-only AI coding was independently replicated, with agreement measured before adjudication. AI identification then used device-specific external public evidence, including official regulatory documentation, manufacturer materials, and scientific sources where applicable. RESULTS: Overall, 386/1546 devices (25.0%) met the explicit device-specific AI criterion. Only 28 (1.81% of all devices; 7.3% of AI-identified devices) contained qualifying AI evidence directly in EUDAMED, whereas 358 (23.2% of all devices; 92.7% of AI-identified devices) required external sources. EUDAMED-only coding showed 98.12% agreement (Cohen's κ=0.318). Class IIa predominated among AI-identified devices (254/386, 65.8%); diagnosis and medical data interpretation accounted for 234/386 (60.6%). CONCLUSION: In this pre-mandatory-use snapshot, EUDAMED alone substantially underrepresents publicly substantiated AI functionality among the software-device records examined. Structured AI metadata, explicit provenance, and lifecycle reporting could improve transparency and regulatory oversight.