AstraZeneca Table Talk with Wayne Filin-Matthews – Global Head of Enterprise Architecture

On September22, 2025, I had the pleasure of participating as the invited guest speaker for the AstraZeneca Table Talk: The Architect’s Edition webinar. The webinar discussion was hosted by AstraZeneca’s Chief Architect, Wayne Filin-Matthews. The webinar discussion focused on AI Ethics from the Architect’s perspective. The live webinar was well attended and there were lively insightful discussions with questions from the people in attendance.

We were able to have deeper discussions around AI Ethics across the AI architecture framework, AI orchestration, having governance at scale, monitoring AI solutions at scale, human-in-the-loop, and human-on-the-loop. One of the most interesting opportunities I had was to get a brief review of AstraZeneca’s AI architecture. Wayne’s team at Astra Zeneca is on the cutting edge of AI architecture. One aspect that contributes to this is Wayne’s propensity to research and taking the time to be plugged-in to what is going on in AI. His encouragement and support for his team to constantly learn because AI is moving fast and constantly changing is also a key aspect of his leadership.

The following represents a summary, key points, and takeaways from the questions I was sent in advance that we dived into as well as ones we did not have a chance to get into.

Summary

AI ethics and governance must be operationalized in biopharma, particularly when deploying Agentic AI in areas like clinical trials, drug discovery, and patient data management. AI ethics and governance highlight’s seven pillars: risk identification, accountability, bias mitigation, data governance, transparency, stakeholder engagement, and continuous monitoring. Each is mapped to concrete biopharma scenarios, such as preventing lab automation errors, ensuring diverse patient recruitment, protecting genomic and EHR data, maintaining regulatory compliance, and enabling explainable drug recommendations.

The central message is that biopharma AI carries high stakes: errors can compromise patient safety, trial integrity, and regulatory compliance. Therefore, safeguards must be embedded from the start through privacy-by-design, fairness testing, explainability, and strict governance aligned with frameworks like NIST AI RMF, ISO 42001, IEEE CertifAIEd™ Ethics Framework, and FDA/EMA guidelines. Responsibility cannot sit with technologists alone; accountability requires cross-functional teams spanning clinical leads, compliance officers, ethicists, and patient advocates.

AI systems must be monitored continuously (monitoring AI solutions at scale) for drift, bias, and malicious manipulation, with kill-switches, immutable logs, and regular audits ensuring trust and regulatory readiness. In short, ethical AI in biopharma means designing for safety, fairness, transparency, and oversight, every step of the AI lifecycle.

Key Points

• Ethical Risk Identification: Biopharma Agentic AI faces autonomy, goal misalignment, bias, privacy, explainability, security, and compliance risks. Examples include lab robots overspending, biased trial recruitment, genomic dataset imbalance, EHR data leakage, and lack of traceability in drug recommendations.
• Accountability Frameworks & Bias Mitigation: RACI roles, autonomy thresholds, and escalation procedures ensure oversight. Bias must be mitigated through dataset diversity, fairness metrics, clinician reviews, and fairness impact statements in FDA/EMA submissions.
• Data Governance, Transparency & Continuous Monitoring: Includes purpose limitation, encryption, provenance tracking, consent management, and vendor restrictions. Transparency through model cards, traceability, plain-language explanations, and pinned versions. Monitoring requires dashboards, drift detection, audits, kill-switches, and immutable logs.

Takeaways

• Ethical safeguards must be embedded at design time: Biopharma AI systems cannot rely solely on post-deployment audits; privacy-by-design, fairness testing, and explainability tooling must be foundational.
• Accountability requires cross-functional roles: Clinical leads, PIs, compliance officers, ethicists, and patient advocates must share responsibility, supported by governance frameworks aligned with both AI (NIST, ISO) and clinical regulations (FDA, EMA).
• Dynamic monitoring and governance are mandatory: Biopharma’s high-stakes AI uses like trial recruitment or drug-response prediction, demand continuous oversight to detect drift, prevent harm, and preserve patient trust.

Wanye and I meet during our MIT AI Leadership Cohort. It was a blast working with him and sharing our AI experience and learning from him and everyone in our cohort. I would like to thank him for inviting me to be his guest speaker. Remember your questions/comments are encouraged and welcomed.