Commission acknowledges implementation challenges and recalibrates high-risk AI obligations for life sciences

The European Commission has unveiled a targeted “Digital Omnibus” proposal on artificial intelligence that could significantly reshape compliance planning for medical device manufacturers, in vitro diagnostics companies and pharmaceutical businesses. Published on 19 November, the package amends elements of the EU AI Act in response to sustained industry concerns that the original timelines and infrastructure were misaligned with implementation realities—particularly in highly regulated health sectors—while preserving the EU’s core safeguards for trustworthy AI.

At its heart, the proposal recognises that high-risk AI requirements cannot bite before the tools to demonstrate compliance exist. Consultations conducted throughout 2025 revealed delays in harmonised standards, uneven readiness of national governance structures and a looming shortage of qualified conformity-assessment capacity. These pressures were most acute for life sciences, where AI embedded in medical devices or diagnostics must navigate a dual compliance pathway under both the AI Act and existing product regimes.

To address this, the Commission proposes a shift away from fixed application dates. Instead, high-risk obligations would apply six months after a formal Commission decision confirms that essential compliance supports—such as harmonised standards, common specifications or guidance—are available. A backstop date of December 2027 would apply for most high-risk systems, while AI that qualifies as, or forms a safety component of, medical devices and IVDs would benefit from a longer 12-month transition following such a decision, with an ultimate deadline of August 2028. The extended runway reflects the complexity of coordinating AI Act requirements with established conformity assessments in medtech.

The Omnibus also tackles a long-standing bottleneck: limited notified-body capacity. A streamlined designation process would allow conformity-assessment bodies to seek approval under both the AI Act and medical-product legislation through a single, consolidated procedure. Existing notified bodies would be given time—up to 18 months from the Omnibus’ application—to build AI competence and apply for designation, reducing the risk of certification gaps just as demand peaks.

Innovation-focused measures feature prominently as well. Real-world testing provisions would be expanded to cover AI systems within regulated medical products, and the regulatory sandbox framework would be strengthened, including plans for an EU-level sandbox by 2028. For pharmaceutical companies, forthcoming guidance on research exemptions is particularly significant, promising clarity on when AI use in pre-clinical research, drug discovery or clinical-trial design falls outside full provider or deployer obligations.

Administrative relief rounds out the package. Simplifications previously reserved for SMEs would extend to small mid-caps, staff AI-literacy duties would shift from mandatory to encouraged, and certain low-impact AI systems would no longer need to be registered in the EU database. Together, these steps aim to reduce friction for companies using AI in ancillary or preparatory functions.

The proposal now enters the ordinary legislative process, with negotiations in the European Parliament and Council expected to run through 2026. While changes remain possible, the Digital Omnibus signals a more pragmatic phase of EU AI governance—one that offers the life sciences sector valuable breathing room without retreating from the bloc’s commitment to safe, human-centric AI.

(Source: Osborne Clarke)