Continuous Oversight: a Key Principle for Implementing AI in Pharmaceutical Manufacturing and Quality Control

Artificial Intelligence (AI) is one of the hottest topics in business right now as executives try to take advantage of the potential benefits. In many respects, the pharmaceutical industry is no different, with the potential of AI in R&D looking particularly exciting. But the pharmaceutical industry is also very different from other industries in a number of areas, especially in relation to manufacturing and quality control.

In pharmaceutical manufacturing and quality control, technologies (i.e., “computerized systems”) can be broadly put into two categories:

• Technologies involved in GMP critical applications
• Technologies not involved in GMP critical applications

GMP (Good Manufacturing Practice) critical applications are those that have a direct impact on patient safety, product quality, and/or data integrity.

Understandably, regulators are most actively focused on AI technologies that are involved in GMP critical applications. Draft guidance has been issued on both sides of the Atlantic as regulators in the US and EU seek to set the parameters within which pharmaceutical organizations must operate as they implement AI technologies.

As a result, clarity is emerging, with continuous oversight being one of the main principles.

AI Defined

AI can be defined in a number of ways. In relation to GMP critical applications in the pharmaceutical industry, it can be most helpful to focus on how technologies, AI or otherwise, get their functionality.

With that in mind, AI technologies are a type of software application (commonly referred to as a model) that gets its core capabilities and functionality through training with data. This sets AI apart from traditional software applications that get their functionality through programming.

More AI Terms and Definitions

AI is a very broad concept, so it’s also important to highlight other terms and definitions relevant to GMP critical applications.

This includes Machine Learning (ML). ML is a subset of AI focused on machines that get their capabilities and functionality from data rather than programming.

Both AI models and ML technologies get their capabilities and functionality from data, but there are two different approaches.

• Static AI models – a static AI model gets its capabilities and functionality from a specific set of data. This knowledge remains fixed.
• Dynamic AI models – a dynamic AI model continuously and automatically learns from new information and changes to its environment. It then uses this new knowledge to adapt its performance.

In GMP critical applications, the intended use of an AI model is also an essential consideration, with a particular focus on outputs. Again, there are two main approaches:

• Deterministic AI models – a deterministic AI model will always produce the same output when given the same input, making it highly predictable.
• Probabilistic AI models – a probabilistic AI model uses probabilities and likelihoods. This means it might not produce the same output when given the same input. As a result, it is a lot less predictable than deterministic AI models.

EU Guidance on Implementing AI in Pharmaceutical Manufacturing and Quality Control

The European Commission has published updates to EudraLex as part of a public consultation (the consultation ends in October 2025). EudraLex contains the EU’s regulations and guidelines for the pharmaceutical industry. The development of new AI technologies and solutions was a key driver for updating these guidelines and launching the consultation.

Chapter 4 – Documentation and Annex 11 – Computerised Systems have both been updated to support innovation in pharmaceutical manufacturing while “ensuring regulatory harmonisation”.

The update to Chapter 4 focuses on the importance of documentation and risk management principles. The revised chapter includes the following:

“The accountability for the integrity of documents, records or (raw) data produced or processed with artificial intelligence or any other automatic means (e.g., validation scripts) rests with the regulated user.”

The update to Annex 11 focuses on enhanced requirements for lifecycle management and continuous oversight of computerized systems, including those that include AI models.

In addition to updating Chapter 4 and Annex 11, a new annex, Annex 22 – Artificial Intelligence, has been introduced. Annex 22 focuses on AI, outlining the requirements for using AI technologies in the manufacture and quality control of pharmaceutical products.

Furthermore, the EU’s new guidelines are very specific about the types of AI models that can be used in GMP critical applications. Specifically, they must be static and deterministic. In other words, only highly controlled and predictable AI models are suitable for GMP critical applications according to EU guidelines, and all models must be subjected to documented and risk-based continuous oversight.

FDA Guidance on Implementing AI in Pharmaceutical Manufacturing and Quality Control

The FDA published draft guidance (Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products) at the start of 2025 on the use of AI in pharmaceutical manufacturing and quality control.

This guidance is less specific about the types of AI models and technologies that can and can’t be used, but there is a similar level of focus on continuous oversight. That focus centers on life cycle maintenance and the importance of ensuring an AI model remains fit for use over the drug product life cycle. This aligns with broader regulatory trends that emphasize a total product life cycle (TPLC) framework with a risk-based validation of AI models.

Conclusion

The publication of new and updated guidance on AI technologies by regulators on both sides of the Atlantic is a welcome development. Continuous oversight and “life cycle maintenance” are essential principles, with concepts such as “human-in-the-loop” being introduced (human-in-the-loop is where human judgment and oversight are intentionally integrated into processes that involve AI technologies).

The guidance will evolve, not least because it is a draft in the US and in a consultation phase in the EU. But the industry now has clearer parameters for exploring the potential of AI technologies to enhance productivity and a range of other performance metrics, including process efficiency, error rates, product quality, throughput, and more.

In future blog posts and whitepapers, my colleagues and I will explore further the topic of AI technologies in pharmaceutical manufacturing and quality control with a specific emphasis on GMP critical applications, compliance, and validation. We will keep you updated on the regulatory guidelines, and our experienced engineers will provide insights on both the technologies that can have a positive impact on your operations and the practicalities of implementing them.

Follow our LinkedIn page to stay up-to-date, and if you have a specific query, please get in touch.

Subscribe to our latest insights