Using AI Without Losing Accountability

Overview

AI does not change where accountability sits in pharmaceutical manufacturing. It changes what is possible — the speed at which documentation can be produced, the depth at which data can be analysed. But the regulatory obligations that determine who is responsible for GMP decisions have not changed. This policy establishes the governance framework that makes those obligations explicit when AI is part of the process.

What this policy covers

• AI risk classification using the FDA two-dimensional framework — decision consequence × model influence
• Pre-deployment requirements: context of use definition, connected document governance, platform assessment
• Human oversight protocol: five-phase obligations for every AI-assisted GMP activity
• Organisational governance: AI system registry, vendor management, training and competency requirements
• Lifecycle management: change control, performance monitoring, vendor update management
• Documentation and dual audit trail requirements for inspection readiness
• Generative AI and LLM-specific controls under EU GMP Annex 22

Who this is for

• VP Quality and Head of Quality at pharmaceutical manufacturers
• QA Directors and Managers responsible for AI governance
• CIOs and Digital Transformation leads deploying AI platforms
• Regulatory Affairs teams assessing AI compliance requirements

Regulatory basis

• 21 CFR 211.22(c) — Quality unit review and approval obligations
• FDA Draft Guidance: AI in Drug and Biological Products (January 2025)
• FDA-EMA Joint Principles for Good AI Practice (January 2026)
• EU GMP Annex 22 — AI in GMP computerised systems (draft)
• MHRA AI Regulatory Strategy (2024)