Inside Annex 22: Guiding the Future of AI in Pharmaceutical Manufacturing
Author: Nicola Rice, Head of Programmes
This article is a follow-up to our recent blog post reporting from the HBA Dublin Chapter event hosted by MSD Swords, where discussions around AI and regulatory expectations generated significant interest.
Following strong engagement and numerous requests for deeper insight, this post takes a closer look at Annex 22 — exploring what it covers, how it fits within the existing GMP framework, and why it matters for teams navigating the use of AI in pharmaceutical manufacturing.
Artificial Intelligence (AI) is transforming how industries operate, from automotive to finance and healthcare. Yet, within the pharmaceutical sector, adoption has been slower. The reason is clear: while AI offers immense potential, its complex and adaptive nature hasn’t easily fit into traditional regulatory frameworks.
That’s now beginning to change. The European Commission’s newly drafted Annex 22 aims to close this gap. For the first time, it sets out clear expectations for the safe and compliant integration of AI and Machine Learning (ML) in Good Manufacturing Practice (GMP) environments.
Annex 22 expands upon Annex 11 (Computerised Systems) and is introduced alongside updates to Chapter 4 (Documentation). Together, these revisions modernise the EU’s GMP framework, creating a harmonised, risk-based model that safeguards patient safety, product quality, and data integrity in an AI-enabled era.
A Modernised GMP Framework
The introduction of Annex 22 is part of a wider digital update to the EudraLex GMP guidance:
• Chapter 4 strengthens controls around digital documentation and traceability, ensuring electronic and AI-generated records meet ALCOA+ data integrity standards.
• Annex 11 has been revised to reflect modern AI and cloud-based systems, aligning directly with Annex 22 and reinforcing expectations around system validation, data security, and human oversight.
Together, these updates provide the regulatory confidence manufacturers need to adopt AI responsibly, building trust through transparency and validation.
What Annex 22 Covers
Annex 22 is the EU’s first AI-specific GMP guideline, defining how static and deterministic ML models can be used in critical manufacturing applications that impact patient safety, product quality, or data integrity.
In practice, this means:
• Only static models, those that do not change during use, are considered suitable for critical GMP use.
• Deterministic models, which produce consistent outputs for identical inputs, are preferred over probabilistic or continuously learning systems.
Dynamic, self-learning models (and large language models like ChatGPT) are therefore currently out of scope for critical manufacturing processes. The focus is on control, consistency, and explainability- principles that align with GMP’s core objective: maintaining product quality and patient safety.
Core Requirements: Trust Through Transparency
Annex 22 defines a clear lifecycle for AI systems, emphasising:
• A defined intended use, describing how and where the model will be applied.
• Comprehensive lifecycle documentation from model design to deployment.
• Explainability of AI outputs- decisions must be scientifically interpretable.
• Risk-based validation, scaled to potential impact on product quality or safety.
• Continuous monitoring for performance, drift, or bias over time.
In essence, Annex 22 transforms AI compliance from an afterthought into a structured, measurable discipline – allowing manufacturers to innovate confidently within a regulated framework.
Harmonised Integration: A Unified Regulatory Model
The real strength of these 2025 EudraLex revisions lies in how Annex 22, Chapter 4, and Annex 11 interconnect. Together, they form a unified framework for digital manufacturing, one that recognises the increasing convergence of automation, analytics, and regulatory oversight.
This harmonised approach allows companies to scale regulatory effort to the risk and criticality of each system, ensuring both agility and accountability in AI deployment.
What This Means for the Industry
These updates signal more than a regulatory shift, they represent a cultural transformation in how pharma views digital technology.
Manufacturers, quality teams, and validation professionals will need to:
• Establish AI governance frameworks that oversee model selection, validation, and monitoring.
• Strengthen risk assessment and human-in-the-loop oversight mechanisms.
• Foster collaboration between QA, IT, and data science teams.
Technology providers and AI developers must also design systems that are validation-ready and capable of demonstrating transparency and explainability, qualities that will soon be central to vendor selection.
Innopharma’s Perspective: From Regulation to Readiness
At Innopharma Technical Services, we see these regulatory updates not as constraints, but as catalysts for progress. They formalise what many leading manufacturers already understand, that innovation and compliance are not opposites but partners in building smarter, safer production systems.
We’re embedding AI literacy and digital skills into our training programmes, preparing the next generation of validation engineers and quality professionals to lead in this new landscape.
By combining education, digital expertise, and regulatory insight, Innopharma is helping the life sciences industry embrace AI with confidence- shaping the future of manufacturing in a digital age.
The Future Starts Now
The release of Annex 22 marks a turning point for the industry, setting a clear, risk-based pathway for responsible AI adoption.
Companies that prepare early, aligning systems, documentation, and skills, will be best positioned to lead in this new era of intelligent, compliant manufacturing.
At Innopharma Technical Services, we’re proud to be part of that journey, bridging innovation and compliance to power the future of digital pharma.
