FDA, Sanofi, Gilead Weigh AI Readiness and Validation Risk at PDA/FDA Conference

At the 2026 PDA/FDA Joint Regulatory Conference, a session on "Advancing Digital Tools and Data Utilization in a CGMP Environment" included presentations on artificial intelligence (AI)-ready manufacturing data along with a panel discussion on validating AI tools under current good manufacturing practice (CGMP). Jack Prior, ScD, head of MSAT process monitoring and data science/AI strategy at Sanofi, presented a data maturity framework for biologics manufacturing, while Liatte Closs, PharmD, MPH, BCSCP, supervisory regulatory specialist at FDA's Office of Manufacturing Quality, examined the foundational quality and data-integrity principles underpinning digital transformation. Both later joined a panel moderated by Christopher Downey, PhD, director of the Division of Pharmaceutical Manufacturing Assessment VI at FDA's Center for Drug Evaluation and Research (CDER), alongside Bob Buhlmann, head of digital quality at Gilead Sciences.¹

What data maturity framework did Sanofi present for biologics manufacturing?

Dr Prior introduced a scoring model, built for Sanofi's manufacturing science and technology (MSAT) network, that rates each process on a scale of 0 to 100 across two dimensions: what data is available and how well it is delivered. On the "what" side, 6 weighted categories cover critical quality attributes at drug substance and drug product, batch output and bioreactor step yields, discrete batch-record and continuous sensor data, genealogy linking raw materials to finished product, and the ability to calculate derived values. On the "how" side, data is scored for freshness, frictionless and broad accessibility, authenticity, standardization, and structured format, he explained.

Achieving that maturity remains difficult, Dr Prior said, given misaligned priorities across quality, digital, and manufacturing functions and legacy systems not built for data science, though he argued the rapidly compounding capability of agentic AI, which he said has been doubling roughly every 3.5 months since 2025, could help close those longstanding data gaps.¹

What foundational elements does FDA say digital transformation requires?

Dr Closs said digitalization is shifting pharmaceutical manufacturing from subjective assessments toward objective, data-driven quality systems but stressed that it does not eliminate human accountability. Predicate rules on validation and record-keeping still apply, and quality-unit oversight remains non-negotiable, she asserted.

Dr Closs outlined 3 foundational elements for evaluating new technology: quality risk management proportionate to a system's risk, consistent with International Council for Harmonisation (ICH) Q9; data integrity, grounded in the ALCOA principle that records be attributable, legible, contemporaneous, original, and accurate, as required under Code of Federal Regulations Title 21, Part 211 (21 CFR 211);² and a "capable facility" that sustains compliance over time rather than merely meeting the minimum standard. She also pointed to the ICH Q10 framework, under which digitally connected process monitoring, corrective and preventive action, and change management can create a self-reinforcing quality improvement loop.¹

What did FDA's warning letters reveal about AI and data integrity risks?

Dr Closs described what she called FDA's first documented warning letter citing inappropriate use of AI in pharmaceutical manufacturing. A firm used AI agents to draft specifications, standard operating procedures, and master production and control records without adequate quality-unit review, violating 21 CFR 211.22(c)² and, by extension, process-validation requirements under 211.100(a).²

"AI and digital tools themselves are not problematic," Dr Closs said, but any AI output must be reviewed and cleared by an authorized member of the quality unit, a responsibility that cannot be delegated, she stressed.¹

In a second case study, FDA investigators at a sterile-manufacturing facility found microbiological plates showing growth on one day and appearing clean days later, with the originals discarded and laboratory forms pre-filled before testing occurred, a violation of 21 CFR 211.194(a).² FDA's suggested remedy in this second case included tamper-resistant, time-stamped digital photographs of test plates to create unalterable, contemporaneous records, Dr Closs explained.¹

How should companies validate AI tools under CGMP?

Asked how to determine which AI use cases require formal validation, Dr Closs reiterated that documentation and validation rigor should be commensurate with the risk of the decision a tool supports. Tools making final decisions reviewed by a quality unit require full CGMP validation, while support tools used case by case, such as for investigations, need only "fit for use" justification that can withstand inspection, she explained.

Gilead’s Buhlmann asserted that companies need an explicit scaling validation process rather than an all-or-nothing approach. Simple tools warrant lightweight documentation, while AI used to build systems like a laboratory information management system warrants more rigor, he said. Dr Prior added that validation should not block root-cause discovery, stating that investigators should be able to pursue root cause with imperfect data before confirming it with validated data. Dr Prior also called for a less binary, more proportional approach to validation industry-wide.¹

The PDA/FDA Joint Regulatory Conference is took place September 14-16, 2026, in Washington, DC.

References

1. Parenteral Drug Association. PDA/FDA Joint Regulatory Conference 2026. Accessed September 14-16, 2026. https://www.pda.org/global-event-calendar/event-detail/pda-fda-joint-regulatory-conference-2026#agenda.
2. Code of Federal Regulations. Current good manufacturing practice for finished pharmaceuticals. 21 CFR §211.22(c), §211.100(a), §211.194(a). Accessed September 16, 2026. https://www.ecfr.gov/current/title-21/chapter-I/subchapter-C/part-211