Introduction
Ask someone working in QA at a sterile manufacturing site what takes up most of the day, and documents will usually come up. Batch records, deviations, CAPAs, change controls, SOPs and audit responses all need careful attention, often against tight deadlines. AI tools are now readily available, and people are already experimenting with them. The question for the Quality Unit is not whether these tools will be used, but how to make sure they are used responsibly.
Used properly, AI can take some of the routine work out of preparing and reviewing information. It cannot replace technical judgement, site knowledge or the accountability that comes with a QA decision.
Where AI Can Help
The most sensible starting point is support work: let AI organise, summarise or challenge information, then have a competent person verify the result.
SOPs and controlled documents: Prepare a first draft, improve readability, identify inconsistent terminology or flag steps that appear to be out of sequence. The process owner must still confirm that the document reflects the actual operation.
Deviation investigations: Organise the event into a timeline, group potential causes under the 6M categories and suggest questions for interviews. This can broaden the investigation and help avoid settling too quickly on ‘human error’.
Trend and history reviews: Summarise previous events involving similar issues, such as glove damage or Grade A environmental monitoring excursions, to help investigators identify recurring patterns.
Change controls: Generate an initial list of potentially affected procedures, systems, equipment and training requirements. QA and the relevant subject-matter experts decide what is genuinely impacted.
Audit and inspection preparation: Generate practice questions on topics such as the Contamination Control Strategy (CCS), Aseptic Process Simulation (APS) and aseptic interventions. Use the questions to test whether explanations are clear and supported by evidence.
Training and guidance: Turn an approved SOP into a short knowledge check or a concise learning aid. AI can also provide an initial overview of new guidance, but the original document remains the reference.
A Practical Example: Grade A EM Excursion
Consider a viable environmental monitoring excursion during aseptic filling. After removing identifying and confidential details, an investigator enters the event description into the company’s approved AI tool. The investigator asks for a timeline, possible contamination routes linked to personnel, material transfers, interventions and airflow, and a list of questions for the operators.
The output may provide a useful starting structure. It does not establish the root cause. The investigator still needs to review the interventions log, gowning records, relevant environmental monitoring data and smoke study evidence, and then assess the facts against the actual process. AI has helped organise the investigation; the investigator remains responsible for the assessment and conclusion.
How to Get Useful Answers
The quality of the output depends heavily on the information and instructions provided. In practice, a few habits make a noticeable difference:
Provide relevant context: Describe the type of site, process and regulatory setting without sharing confidential information.
Be specific about the task: Ask for a timeline, a risk-based question list or a one-page summary rather than simply asking the tool to ‘review’ something.
Ask it to identify gaps: Request a list of missing facts, assumptions and points that need confirmation.
Challenge the response: Treat the first answer as a draft. Ask follow-up questions and test whether the suggested reasoning is supported by the information supplied.
Where the Line Must Be Drawn
There are boundaries that should not be negotiable in a GMP environment:
No AI-made GMP decisions: Batch disposition, QP certification, final root-cause conclusions, CAPA approval and risk acceptance remain with authorised, accountable personnel.
Protect confidential information: Batch details, product formulas, patient information and unpublished results must only be entered into tools approved for that data by the company.
Verify every reference and claim: AI can produce convincing but incorrect statements, including fabricated guideline clauses or publications. Check references against the original source and confirm technical statements independently.
The signatory remains accountable: If AI-assisted wording is included in a GMP record, the person reviewing and approving it owns the content. Applicable data-integrity expectations, including ALCOA+, still apply.
Common Traps
The most obvious risk is copying and pasting an answer without checking it. AI may introduce the wrong room grade, a product name that does not apply to the site or a procedural step that is not part of the process. A second risk is generic writing: a deviation report or CCS section that could describe almost any facility is unlikely to demonstrate a sound, site-specific assessment. Finally, fluent language should not be mistaken for sound reasoning. If staff are using AI, it is better to discuss that openly, within company rules, so that useful practices and errors can be shared.
Regulatory Perspective
The European Commission published draft revisions to Annex 11 and Chapter 4, together with a draft Annex 22 on artificial intelligence, in July 2025. The consultation closed on 7 October 2025. As of November 2026, Annex 22 remains a draft and is not in force; its status should be checked before the article is used for regulatory decision-making.
The direction of travel is clear: systems used in GMP activities need an appropriate level of control, and the intended use matters. Generative AI and chat-based tools are best treated as support for non-critical tasks, with human review, rather than as decision-making systems for critical GMP activities. The EU AI Act also places importance on AI literacy for people using AI systems. In short, AI may help QA prepare and assess information, but it does not take responsibility for the outcome.
A Sensible Way to Start
A Quality Unit does not need a complicated AI programme to begin. It can start by finding out which tools employees are already using and for what purpose. The next steps are to distinguish support activities from uses that could affect product quality, patient safety or data integrity; provide an approved tool; define what information may be entered; and explain how outputs must be checked. Training should use realistic site examples, including examples where the AI response was wrong. The guidance should then be reviewed as regulatory expectations develop.
Final Thought
AI will not replace the Quality Unit. It may, however, change how some of its work gets done. Less time spent drafting routine text could mean more time reviewing evidence, understanding process behaviour, supporting investigations and speaking with operations. Those activities are where QA adds value.
The test is straightforward: before using AI-assisted content in a GMP record, ask whether you understand it, whether you have verified it and whether you are prepared to sign it. If the answer to any of those questions is no, it is not ready.
About the Author

Varadharaj Vijayakumar is a pharmaceutical industry professional with over 17 years of experience in pharmaceutical and biopharmaceutical manufacturing. His expertise spans aseptic processing, sterile manufacturing, contamination control, process validation and EU GMP Annex 1 implementation. He is passionate about advancing sterile manufacturing practices, sharing practical industry knowledge and exploring how emerging technologies can strengthen pharmaceutical quality and operational excellence.
Disclaimer: The views and opinions expressed in this article are those of the author and do not necessarily reflect the views or positions of Quascenta.
