AI Regulatory Milestones: Qualification & Observatory

First qualification opinion on AI methodology

In March 2025, EMA's Committee for Human Medicinal Products (CHMP) issued a qualification opinion accepting clinical trial evidence generated by an AI tool supervised by a human pathologist. This was the first time EMA considered data generated with AI assistance to be scientifically valid for regulatory purposes.

The AIM-NASH tool helps pathologists analyse liver biopsy scans to determine the severity of metabolic dysfunction-associated steatohepatitis (MASH, formerly NASH). It helps researchers obtain clearer evidence on treatment benefits in clinical trials with fewer patients.

Key takeaway: AI can support — not replace — expert clinical judgement. Human supervision remained essential for regulatory acceptance.

AI Observatory report (2024)

The first annual report of the European medicines regulatory network's AI Observatory compiles experience from 2024 in:

• Enhancing productivity
• Automating tasks
• Supporting data-driven decisions across the medicine lifecycle

A horizon scanning report reviewed scientific literature and EU-funded projects to identify gaps, challenges, and opportunities for integrating AI in medicine regulation. The observatory captures and shares experiences to inform the Network Data Steering Group's multi-annual AI workplan.

AI tools in regulatory practice

EMA extended its AI-enabled knowledge mining tool Scientific Explorer to help assessors search regulatory scientific information efficiently — including public assessment reports for initial marketing authorisation applications. The tool supports decision-making via efficient retrieval of regulatory precedents.

Relevance for Hong Kong healthcare professionals

These milestones illustrate a global trend: regulators are moving from theoretical AI guidance to validated use cases and structured monitoring. Hong Kong clinicians and researchers should expect increasing scrutiny of AI-generated evidence in trials and publications, while also recognising legitimate productivity uses — provided human oversight, documentation, and institutional governance are in place.

Sources: EMA — Artificial intelligence; AIM-NASH qualification opinion