EU GMP Annex 11 & Annex 22: The New Compliance Reality for AI in Pharma

Pharma is in the middle of its biggest regulatory shift in more than 30 years. For the first time, regulators are explicitly addressing AI, with dedicated guidance, clearer governance expectations, and new design constraints.

Why this reform is different

Regulatory updates happen all the time. So why does this one deserve special attention?

The scope of the revision is broader than anything seen before in GMP. It explicitly addresses AI model behaviour, cloud accountability, data governance at the quality system level, and the full lifecycle of computerized systems. Several shifts stand out.

Accountability does not transfer to the cloud. Even if systems are hosted, built, or operated by vendors, responsibility remains with the regulated company. Organisations must still be able to explain system behaviour, provide evidence from their own environment, and demonstrate control during inspection.

Data governance is moving beyond individual systems. It is becoming a quality system responsibility. Data integrity is no longer just an IT concern, it sits squarely within the pharmaceutical quality system.

Both Annex 11 and Annex 22 are post-consultation and in consolidation as of early 2026, with finalization expected during 2026. But the direction is already clear. Companies that are waiting for the final wording before acting are already behind. 

Annex 11: The foundation that's often underestimated

Annex 11 is the structural base for everything that follows, including AI. The revised draft expands significantly from the original 2011 version. It now covers full system lifecycle control: supplier management, identity and access control, cybersecurity, backup and recovery, audit trails, archiving, and periodic review.

What used to live in the domain of "good IT practice" is now codified as GMP obligation. That matters because AI never enters a regulated environment on its own. It is embedded in systems, workflows, and cloud infrastructure. If those foundations are weak, model quality becomes irrelevant. The organisation already has a governance problem.

Another important shift is lifecycle thinking. Validation is no longer a one-time event. It is continuous assurance that systems remain controlled, documented, and reviewable over time. This applies across GxP domains, including GMP, GCP, GLP, and GDP. The underlying expectation is consistent: patient safety, product quality, and data integrity must be demonstrable throughout the system lifecycle.

Annex 22: Validating intelligence, not just software

Annex 22 builds on Annex 11 and introduces something new: formal expectations around AI behaviour.
For the first time, regulators are focusing not only on systems, but on model outputs, model behaviour, and model risk. This shifts validation from “does the system work as intended” to “can we trust the model’s output under defined conditions”.

Three characteristics become central:

• Predictable behaviour under defined inputs
• Auditable decision pathways
• Explainable outcomes that can be justified in inspection
  

Without these, AI use in GMP environments becomes difficult to defend.

Annex 22 is not separate from Annex 11. It depends on it. The two frameworks are meant to operate together: one governs the system, the other governs the intelligence within it.

The design constraint most organizations haven't internalized yet

A key point in the current draft is that critical GMP use cases are expected to rely on static, deterministic models.

That has direct implications:

• The model does not update or self-learn once released into production
• Given the same input, it produces the same output, every time
• Any change to the model must go through formal change control and trigger a revalidation assessment
• Generative AI and large language models cannot serve as autonomous decision-makers in critical GMP processes

This creates a clear boundary for AI in regulated environments. Generative AI and large language models may still be used, but typically in non-critical or supportive roles. The starting point is always the same: intended use and risk level define the acceptable technology, not the other way around.

The three principles that haven't changed

Despite the technological shift, core GMP principles remain stable.

Quality must be designed in. Controls cannot be added after deployment. AI systems need governance embedded from the start.

Retrospective validation is not acceptable. A system already in use without a proper validation package is not made compliant by documentation after the fact.

The full lifecycle must be controlled. This includes monitoring, change control, revalidation triggers, and decommissioning. These elements must be defined before go-live.

Inspectors will always return to three questions: is the system fit for intended use, is it controlled throughout its lifecycle, and is it at least as safe and reliable as the process it replaces.

Where to start

The most common mistake is starting with technology selection instead of foundations.

The first step is assessing whether the existing digital and quality systems are mature enough to support AI. Weak governance, fragmented data, or inconsistent validation practices will not improve with AI, they will scale with it.

The second step is defining intended use with precision: process boundaries, decision points, inputs, outputs, acceptance criteria, and human roles. That definition drives everything else, from validation scope to system design and governance structure.

The regulation is not positioned to restrict AI. It is defining the conditions under which AI can be trusted in regulated environments. Those conditions are achievable, but only if the sequence is correct.

If you're trying to figure out what to do with all of this, here's the honest answer: start with your foundations, not with your AI ambitions. Before selecting a model or a vendor, ask whether your underlying digital environment is mature enough. If those fundamentals are shaky, AI will inherit those weaknesses, and amplify them.

Want to know where you stand?

We built the GxP AI Readiness Checklist specifically for this moment: 50 structured questions covering regulatory compliance, validation, human-in-the-loop controls, auditability, data quality, cybersecurity, and more. It's a practical self-assessment, but it's a solid way to see where your gaps are before they surface during an inspection.

Download the GxP AI Readiness Checklist

Understanding Annex 11 and Annex 22 is the first step. The harder question is what this means in practice: Where can AI genuinely support regulated processes? Where does it need to stop? And how do you design AI workflows that stay inside GxP boundaries?

Read next: What Pharma Can (and Cannot) Do with AI Under Annex 22