Cell & Gene Covers the Session, Implement and Manage Risks for AI solutions in cGMP, to be Presented at the 2026 ISPE Annual Meeting & Expo
16 September, 2026
As pharmaceutical manufacturers explore ways to integrate artificial intelligence into regulated operations, industry leaders are emphasizing that successful adoption depends as much on governance and risk management as it does on technology. These themes will be a key focus at the 2026 ISPE Annual Meeting & Expo, where Pfizer's Director of Engineering Validation, Arif Azad, will discuss practical approaches for deploying AI solutions in Good Manufacturing Practice (GMP) environments while maintaining compliance, data integrity, and patient safety.
According to Azad, the pharmaceutical industry is moving beyond AI pilot programs and beginning to evaluate how artificial intelligence can support routine quality and manufacturing operations. Potential applications include batch record review, GxP documentation generation, maintenance signal detection, deviation trending, and periodic quality reviews. However, he cautions that organizations must ensure these tools are implemented within established GMP frameworks rather than treating them as standalone technology projects.
A major distinction highlighted by Azad is the difference between real-time and offline AI applications. Real-time AI systems analyze live or near-live manufacturing data and can identify issues such as equipment drift, documentation gaps, or unusual batch record entries while a process is still underway. By contrast, offline AI focuses on historical data analysis, such as reviewing completed investigations or conducting periodic trend assessments. While both approaches offer value, real-time applications carry significantly greater compliance expectations because they may influence operational decisions that affect product quality. As a result, companies must maintain clear controls around data sources, model versions, audit trails, human oversight, and monitoring processes.
Moving AI from pilot programs to routine GMP use requires what Azad describes as a disciplined lifecycle approach. Organizations should begin by defining the intended use of the AI system, assessing its GxP impact, establishing acceptance criteria, and testing performance against representative production records. Most importantly, initial implementations should maintain a “human-in-the-loop” structure, where AI drafts, flags, prioritizes, or summarizes information while qualified personnel retain responsibility for final decisions. Successful pilots may demonstrate technical feasibility, but routine GMP deployment requires robust controls governing data management, change control, exception handling, ongoing monitoring, and eventual system retirement.
One of the most significant challenges facing manufacturers is underestimating the risks associated with AI implementation. Azad noted that organizations often focus heavily on model performance while overlooking other critical considerations such as data quality, cybersecurity, supplier oversight, model drift, validation requirements, and accountability frameworks. He also pointed to a growing cultural challenge within organizations: some employees place too much trust in AI outputs, while others reject the technology altogether. In his view, neither approach is appropriate. Instead, AI should be managed using risk-based principles similar to those outlined in ICH Q9, with the level of oversight increasing as the technology’s impact on quality decisions becomes more significant.
Another key issue is model bias and transparency. In GMP settings, AI systems must be tested across multiple product types, manufacturing sites, data formats, and known failure scenarios to ensure consistent performance. Verification activities should include analysis of false positives and false negatives, confidence thresholds, explainability assessments, and documented human review processes. Rather than attempting to prove perfection, companies should focus on demonstrating that AI delivers consistent and controlled performance for specific intended uses.
The rise of generative AI also presents new challenges for data integrity. Azad recommends extending traditional ALCOA+ principles by incorporating AI interactions into the broader record lifecycle. This includes maintaining records of prompts, source data, model versions, outputs, timestamps, reviewer actions, and final human decisions. AI-generated content should be treated as a draft until verified by qualified personnel, ensuring that accountability remains clearly assigned and auditability is preserved. If regulators request evidence supporting a decision, organizations must be able to demonstrate what the AI recommended, what changes humans made, and why those changes occurred.
While regulators such as the US Food and Drug Administration and European Medicines Agency have begun establishing principles for AI governance, Azad noted that manufacturers continue to seek greater clarity surrounding dynamic models, generative AI applications, model drift, cloud-hosted tools, supplier oversight, and regulatory notification requirements. Continued global harmonization of expectations will be essential for increasing industry confidence and accelerating adoption.
For companies just beginning their AI journey, Azad recommends starting with lower-risk applications such as deviation analysis, exception triage, maintenance monitoring, and draft review support. Conversely, organizations should avoid attempting autonomous batch release decisions or unrestricted AI-generated GMP documentation until governance frameworks and institutional experience have matured.
Looking ahead, Azad envisions a future where AI becomes a standard component of pharmaceutical quality and validation systems, with tools that are inventoried, risk-assessed, validated, and continuously monitored. If implemented responsibly, AI could improve efficiency, consistency, and insight across manufacturing operations. However, failure to maintain adequate controls could result in data integrity issues, regulatory findings, and loss of trust in digital transformation initiatives. These considerations are expected to be central discussions at the 2026 ISPE Annual Meeting & Expo, where industry leaders will examine how to balance innovation with the rigorous compliance standards that define pharmaceutical manufacturing.
Learn More About the 2026 ISPE Annual Meeting & Expo
About Cell & Gene
Cell & Gene delivers in-depth content about cell and gene therapies from experienced professionals in the field. This publication focuses exclusively on this segment of the life sciences industry.