At the 2026 PDA/FDA Joint Regulatory Conference, representatives from the US Food and Drug Administration (FDA), Sanofi and Gilead Sciences discussed how pharmaceutical manufacturers can assess manufacturing data readiness for artificial intelligence (AI), maintain data integrity and apply risk-based validation to digital tools used in current good manufacturing practice (CGMP) environments.

The session, “Advancing Digital Tools and Data Utilization in a CGMP Environment,” covered data maturity in biologics manufacturing, quality-unit oversight and the appropriate level of validation for AI applications. The conference took place in Washington, DC, from 14 to 16 September 2026. The discussion was reported by BioPharm International.

Sanofi Presents a Data Maturity Scoring Framework

Jack Prior, head of manufacturing science and technology (MSAT) process monitoring and data science/AI strategy at Sanofi, presented a framework that scores biologics manufacturing processes on a scale of 0 to 100. The model evaluates two dimensions: the data available and the way that data is delivered.

The data-availability assessment covers six weighted categories, including critical quality attributes for drug substances and drug products, batch output and bioreactor yields, batch records and continuous sensor data, raw-material genealogy, and the ability to calculate derived values. The data-delivery assessment considers freshness, accessibility, authenticity, standardisation and structure.

Prior identified legacy systems and competing priorities across quality, digital and manufacturing teams as obstacles to improving data maturity. He also discussed the potential of agentic AI to help address persistent data challenges.

FDA Emphasises Human Accountability and Data Integrity

Liatte Closs, a supervisory regulatory specialist at FDA’s Office of Manufacturing Quality, stressed that digital transformation does not remove existing CGMP obligations or transfer accountability from qualified personnel to AI systems. Quality-unit oversight remains central to document approval, process validation and record governance.

The discussion highlighted three foundational elements for digital transformation in a CGMP environment:

• Quality risk management: evaluating digital technologies in line with the principles of ICH Q9
• Data integrity: maintaining records consistent with ALCOA principles and applicable requirements under 21 CFR Part 211
• Sustained compliance: maintaining a capable facility that continues to meet CGMP requirements over time

FDA Warning Letter Highlights Risks of AI-Generated GMP Documentation

Closs also discussed an FDA warning letter issued to Purolea Cosmetics Lab on 2 April 2026. The agency documented that the company had used AI agents to prepare drug product specifications, procedures and master production or control records without adequate quality-unit review. FDA identified this as a violation of 21 CFR 211.22(c). The inspection also found that the company had failed to conduct required process validation before distributing its products, contrary to 21 CFR 211.100.

FDA stated that AI-generated documents used in CGMP activities must be reviewed for accuracy and compliance and cleared by an authorised human representative of the quality unit. The warning letter illustrates that responsibility for CGMP compliance remains with the manufacturer, regardless of whether AI tools are used to support documentation or other activities.

As discussed in Pharmaceutical Technology, manufacturers should verify AI-generated content against approved source materials, involve qualified subject-matter experts in technical review, retain evidence of changes and review, and route final documents through established document-control and Quality Unit approval procedures.

The conference also covered a separate case involving microbiological testing records at a sterile manufacturing facility. According to BioPharm International’s report, investigators found inconsistencies in test-plate evidence and laboratory documentation. The discussion highlighted the importance of contemporaneous records and the potential use of tamper-resistant, time-stamped digital photographs to support data integrity.

Industry Panel Calls for Risk-Based AI Validation

A panel moderated by Christopher Downey of FDA’s Center for Drug Evaluation and Research (CDER) brought together Closs, Prior and Bob Buhlmann, head of digital quality at Gilead Sciences, to discuss practical approaches to AI validation.

The participants supported scaling validation activities according to the risks associated with an AI tool and the decisions it supports. AI applications that make or directly support final quality decisions require rigorous CGMP validation. Tools used to assist case-specific investigations may be assessed through a documented, fit-for-use approach appropriate to their intended purpose and risk.

Buhlmann argued for a scalable validation process rather than an all-or-nothing model, with documentation requirements proportionate to the complexity and criticality of each application. Prior also noted that exploratory data analysis during root-cause investigations should not be unnecessarily constrained, provided that conclusions are subsequently substantiated using appropriate, reliable evidence.

The discussion highlighted the distinction between using AI to support scientific investigation and relying on AI-generated outputs for regulated quality decisions. Existing CGMP requirements, appropriate validation and human oversight remain essential when manufacturers introduce AI into pharmaceutical manufacturing and quality systems.