Agenda

This interactive workshop will put participants into the center stage: We will explore supplier collaboration in a combined AI and GxP context. Participants can decide whether they take the position of the regulated company or the supplier, receive a briefing, prepare their strategy, and come together in a supplier assessment-like situation. In the plenum, teams will share their experiences from the workshop and from real-life situations.

Objectives of this workshop include:

• Identify relevant areas of GxP supplier assessments in an AI context that go beyond usual supplier assessment methods
• Learn what to consider for an AI-related supplier assessment from a supplier perspective
• Experience the necessity of a common language and harmonized life cycle approaches, and their relation to horizontal regulation like the EU AI Act
• Learn what level of experience is needed from suppliers to gain trust in their services and software products
• Learn about good practice guidance to establish AI supplier assessment frameworks

Workshop facilitators will help the groups to prepare and interact, while the session will allow room for exchange of insights through the workshop and from real-world settings. The workshop presents a practical, thought-provoking experience providing a highly relevant framing to the main part of the ISPE AI in Life Sciences 2026 Summit – powered by GAMP®.

The life sciences industry faces a fundamental collision: the unprecedented velocity of AI-driven discovery meets the rigors of traditional GxP validation. With this challenge, we can no longer treat security as a perimeter to be defended; we must treat it as a dynamic governance layer that enables trust at scale. This keynote outlines a blueprint for "validated agility," demonstrating how the Secure AI Framework (SAIF) provides the technical foundation for the FDA’s emerging risk-based credibility guidances. We will move beyond the hype to address the practicalities of maintaining model integrity, managing international data sovereignty in multinational trials, and architecting the "Total Product Lifecycle" approach now expected by global regulators. Attendees will learn how a security-first strategy not only protects data but also accelerates the path to the patient.

The Change Control process in the pharmaceutical industry is an essential component of Quality Management Systems (QMS), ensuring that all changes to processes, systems, equipment, or documentation are systematically evaluated, approved, and implemented in accordance with regulatory standards such as GMP, GAMP5, and ICH Q10. Integrating Artificial Intelligence (AI) into this process can greatly improve decision-making, efficiency, compliance, and risk management. The use of Artificial Intelligence in pharmaceutical change control provides tremendous opportunities to increase productivity, compliance, and product quality. Pharmaceutical companies that adopt AI-driven change management will be better positioned for regulatory success and operational excellence. Documentation must cover AI algorithm qualification, risk assessment, and change impact evaluation.

Pfizer’s Global Regulatory Strategy (GRS) is a leader of AI adoption across Pfizer and the industry   . Our approach begins with a clear governance framework and a compelling vision that prioritizes people—delivering relief, recognition, and relevance by saving time, enhancing impact, and advancing careers. GRS has built one of Pfizer’s most active Communities of Practice, engaging 800+ colleagues with daily AI insights and practical applications. To drive adoption, we embed AI into performance goals, bi-weekly AI office hours assistance, and encourage colleagues with Copilot licenses to dedicate 5% of their time to experimentation and project-creating hands-on learning opportunities that accelerate skill-building. We've developed role-specific training that includes leadership skills for everyone to help drive an AI-ready organization, and we’re redesigning incentives through AI-driven performance metrics to reward innovation and automation. Together, these initiatives are transforming how enterprises prepare for an automated future. Over 60% of colleagues with Copilot licenses use AI daily, saving two hours per week, and GRS ranks in the top 15% for Copilot usage across Pfizer. This session will share actionable strategies for scaling AI responsibly, overcoming cultural resistance, and enabling transformation through governance, collaboration, and innovation.

Unlock the Power of AI: Real Results, Real Impact—See How Eli Lilly is Transforming Quality!

In this session, discover how Eli Lilly’s Quality Organization is harnessing AI to drive meaningful transformation. As digital innovation accelerates, building AI literacy and embedding knowledge management are essential for long-term success. This presentation highlights Lilly’s strategic approach to empowering Quality professionals, from foundational awareness to advanced fluency, ensuring teams are equipped to lead confidently in an AI-enabled future. Attendees will learn how AI is seamlessly integrated into daily quality operations, turning knowledge into action and enabling smarter decisions, faster cycles, and stronger outcomes. Real-world use cases, including automated validation documentation and AI-powered regulatory intelligence, will demonstrate measurable impact and replicable strategies. Participants will leave with actionable insights on how to empower their Quality teams, embed AI into workflows, and solve real-world challenges with scalable solutions. Whether you're beginning your AI journey or seeking to accelerate adoption, this session offers a practical roadmap for transforming Quality through innovation.

This session—presented from the perspective of a Director of Strategy, Global Biologics Operations—lays out a pragmatic, five-year plan to enable GenAI across a large-cap pharma network. We’ll present a capability ladder that begins with secure, validated chatbots for day-to-day knowledge access, progresses through progressive model training and supervised fine-tuning, and opens enterprise Copilot/agent-builder platforms for advanced automation. The talk ties functional use-cases (communication assistants, project coordination copilots, tech-transfer accelerators, predictive maintenance, and yield optimization pilots) to a governance funnel that channels citizen-developer efforts into enterprise-grade solutions with light-touch control and GxP alignment. Attendees will see a repeatable funnel methodology for ideation to rapid prototyping to cost/benefit screening to prioritization, with a bias for quick wins that deliver measurable downtime reduction, yield improvement, and cycle-time savings.

AI is rapidly transforming pharmaceutical manufacturing, offering new levels of insight and automation. However, its non-deterministic nature, where outputs can vary with the same input data, context, and model setup, poses challenges for an industry built on precision and quality. What if the greatest risk isn’t adopting AI—but failing to harness its full potential? As AI redefines what’s possible, we question: How do we unlock breakthrough innovation while safeguarding the highest standards of quality and safety? This talk introduces the concept of “Pharma-Trusted AI,” with examples of where AI can have the most profound impact beyond traditional automation and analytics. We share realigning thinking to see AI as a trusted partner in the efficient delivery of safe, timely medicines where and when they are needed. We'll discuss guardrails for designing AI systems that align with pharmaceutical quality standards, including explainability, validation frameworks, and human-in-the-loop oversight. Along with partner companies, we outline how this is used in pharmaceutical process development for DoEs and for manufacturing operations. Attendees will gain insights into balancing innovation with compliance and productivity, and how to move from risk-avoidance to using AI as a catalyst for continuous improvement while maintaining the integrity and safety that the industry demands.

One of the richest sources for data-driven quality decision making in pharmaceutical manufacturing comes from the operational data generated by automated equipment. While challenges exist in making raw machine information accessible for broader use, integration of AI-enabled platforms opens the possibility of understanding a user's request in natural language to surface data, as well as unique data insights. As workforce challenges continue to grow for manufacturing facilities, AI tools can better onboard and enable personnel. To achieve this goal, manufacturers need to consider laying the building blocks for scalable AI adoption in the regulatory space, including:

• Strategies for creating well-defined, equipment-level datasets needed for AI reasoning. This includes the use of standard libraries, centralized data hierarchies, and architectures compliant with data integrity principles.
• Frameworks for AI platform governance, including human-in-the-loop checkpoints along the data pipeline, approaches to model validation, and human oversight of AI-derived outputs.
• Outlines for how life sciences companies can create their own roadmap to evaluate and adopt artificial intelligence.

This presentation will delve into a compelling real-world case study, showcasing the transformative power of agentic AI in automating the critical processes of scanning, interpreting, and disseminating regulatory updates. We will explore in detail how these intelligent systems can significantly enhance efficiency and accuracy in compliance management. The presentation will discuss innovative strategies for effectively leveraging AI as a unified platform. This approach aims to not only streamline and improve global team collaboration across various departments and geographies but also to revolutionize the way compliance workflows are managed.

A key focus will be placed on sharing the challenges encountered and the best practices developed during the transition from a traditional, reactive compliance model to a forward-thinking, proactive, and AI-driven strategy. Attendees will gain valuable insights into overcoming common obstacles and implementing successful AI solutions in their own organizations. Ultimately, this presentation will foster an open dialogue about the challenges, successes, and critical questions that arise as we collectively endeavor to build a smarter, more intelligent, proactive, and collaborative compliance culture. We invite participants to engage in a discussion that will shape the future of compliance in an increasingly complex regulatory landscape.

Artificial Intelligence is rapidly evolving from experimental digital support to an operational capability within Regulatory Affairs and CMC functions. As organizations begin embedding AI into authoring, data verification, and lifecycle management activities, regulators are placing increasing emphasis on governance, validation, and traceability to ensure output remains reliable, transparent, and submission ready.

This presentation will provide a practical, regulatory-focused perspective on how organizations can transition from fragmented AI tools to structured governance models aligned with FDA, EMA, and emerging global expectations. It will explore how Regulatory and CMC teams can responsibly integrate AI into workflows while preserving data integrity, inspection readiness, and confidence in submission quality across biologics and advanced modalities.

Attendees will gain a clear understanding of how AI-enabled processes, when supported by robust governance, can strengthen regulatory consistency, improve lifecycle control, and support inspection-defensible submissions.

Learning Objectives:

• Understand the evolving regulatory expectations for AI use within Regulatory Affairs and CMC environments.
• Learn the governance components necessary to ensure AI-supported output remains compliant, transparent, and submission-ready.
•    Identify practical strategies for integrating AI into regulatory operations while maintaining data integrity, traceability, and inspection readiness.

The adoption of AI in life sciences requires not only robust technical foundations but also a regulatory- and quality-centered framework for governance, inspection readiness, and risk management. This session, co-presented by Redica Systems and Gilead Sciences, will demonstrate how AI-driven risk assessment can strengthen compliance strategies while safeguarding product quality, patient safety, and data integrity. The first part will introduce Redica’s risk assessment methodology, which evaluates multiple “risk pillars"—including quality signals, regulatory history, and manufacturing complexity—to provide a holistic view of site and CMO risk. AI models will be described that label observations, structure unstructured text data, and enable comprehensive workflow evaluation. Attendees will learn how AI is trained, deployed, and validated to support proactive compliance decision-making and inspection preparedness.

The second part, presented by Gilead Sciences, will show how a sponsor company integrates internal operational data with external intelligence from the Redica platform to create a unified quality risk framework. Leveraging advanced AI tools, Gilead will illustrate how this dual-source approach enables rapid, actionable insights while maintaining scalability and reusability. Together, these perspectives provide an end-to-end view of AI-enabled risk management, from methodology design to sponsor implementation, highlighting lessons learned and strategies to scale responsibly across GxP environments.

Modern quality assurance demands perfection at scale - reviewing massive volumes of documentation while catching every error that could impact patient safety or regulatory standing. It's a nearly impossible standard for human reviewers alone. This session explores how one biomanufacturing facility augmented its quality team with AI technology, creating a partnership where machines handle exhaustive pattern detection while humans provide context, judgment, and decision-making. This presentation shares practical experience from deploying AI as a quality advisor in a biomanufacturing facility. We'll demonstrate how Axella Bio integrated AI-powered error detection into their existing QA workflows without disrupting established processes or requiring expensive system replacements. The results tell a compelling story of improved error detection rates, faster batch review cycles, and quality teams spending more time on meaningful analysis rather than page-by-page checking.

Artificial intelligence (AI) and Generative AI (GenAI) are transforming information management by streamlining workflows, enhancing compliance, and accelerating decision-making. This presentation outlines a multi-pronged strategy for integrating AI into regulatory frameworks, emphasizing collaboration, standards, and scalable infrastructure. Quality (CMC) information management and Module 3 will be used as a representative example.

The approach focuses on harmonizing data standards, automating authoring and regulatory workflows, and fostering partnerships among industry, regulators, and technology providers. It highlights the importance of infrastructure modernization and organizational readiness as well as the foundational need of preparing the data that is AI-ready (standardized, structured, and interoperable) and aligned with key frameworks such as IDMP and HL7 FHIR, and guidelines like ICH M4Q(R2) and ICH M16.

Despite growing interest, AI adoption remains uneven due to fragmented data environments, inconsistent practices, and challenges in integrating structured and unstructured data. Organizations must navigate evolving technological capabilities, regulatory expectations, and compliance considerations. A well-orchestrated, forward-looking strategy and cross-sector collaboration are essential to overcome adoption barriers and unlock the full potential of AI in regulatory and overall life sciences.

The Operations Quality Assurance Innovation Accelerator is a cross-functional, data-driven initiative designed to transform quality processes in the life sciences industry. By integrating governance, leadership, and innovation, it empowers teams to tackle regulatory and operational challenges through collaborative intelligence and emerging technologies, such as AI and automation. The program accelerates idea development and implementation via structured intake, collaborative workshops, and transparent prioritization, fostering both innovation and compliance. Its key pillars include tiered governance for AI alignment, leadership empowerment, project incubation for automation and data visibility, and responsible experimentation with AI tools. The Accelerator helps address manual, fragmented operations that raise compliance risk and slow innovation, focusing on improving automated assessments, reducing compliance observations, and increasing process efficiency and timeliness. Beneficiaries include quality, regulatory, and operational teams across affiliates, business SMEs, CoE leads, site representatives, and ultimately patients through improved compliance and product delivery. Offering robust, scalable AI-enabled processes, the program delivers measurable value—cost avoidance, time savings, and increased throughput—and lays a foundation for continued transformation, ensuring the organization effectively adapts to evolving regulatory and technological landscapes.

AI in pharmaceutical manufacturing is outpacing the guidance designed to govern it. Regulators are actively wrestling with a fundamental question: do AI applications require new frameworks, or can existing guidance be intelligently mapped onto them? This panel brings together regulatory perspectives to stress-test both approaches, with a specific focus on manufacturing and CMC applications, where the stakes for patient safety and compliance are highest.

Connect with industry leaders at our Welcome Reception. Enjoy food and beverages in the Expo Hall while networking with exhibitors and attendees.

As AI reshapes industries, mastering its use in regulated environments is becoming critical. Digital solutions are regulated across pharma and beyond, often creating a mental barrier to adoption. This keynote shows how to move beyond that mindset and build a culture that fosters AI and digital innovation.

The convergence of digital innovation and regulatory science presents a transformative opportunity to accelerate patient access to medicines by fundamentally reimagining the global regulatory ecosystem. Traditional regulatory processes—characterized by paper-based submissions, regional silos, staggered submissions and sequential reviews—have historically delayed approvals, increased regulatory complexity, and limited timely access to critical medicines around the world. Emerging 21st century technologies, including cloud computing, artificial intelligence, and standardized data frameworks such as HL7 FHIR and ISO IDMP, are enabling a paradigm shift toward a fully digital, interconnected regulatory environment. At the core of this transformation is the implementation of structured, standardized data at the source, seamlessly integrated across sponsor systems and exchanged through cloud-based platforms. This architecture enables real-time, parallel regulatory review, enhanced transparency, and advanced analytics, replacing fragmented workflows with a collaborative, data-driven model.

The impact on patients is profound. Case studies leveraging cloud-based regulatory platforms demonstrate reductions in global lifecycle approval timelines from over four years to less than one year—and often under six months—dramatically accelerating access to innovative therapies to patients around the world in record time. Simultaneous global submissions and reliance-based pathways enable regulators worldwide to access and review the same dossier in real time, resulting in approval rates up to 800% to 2700% faster compared to traditional approaches. These efficiencies not only expedite patient access but also enhance supply chain agility, increase manufacturing capacity, and reduce medicinal waste. Ultimately, the integration of digital technologies into regulatory ecosystems shifts the focus from process-driven constraints to patient-centered outcomes. By enabling faster, more predictable, and globally harmonized decision-making, this new model ensures that life-saving medicines reach patients sooner, regardless of geography, marking a critical advancement in public health.

This session explores how Generative AI and multi-agent systems can support GxP audit preparation by addressing a longstanding challenge in Quality: fragmented insights. Compliance data—such as SOP changes, CAPA trends, regulatory updates, and supplier history—often sit in silos. This session introduces a conceptual “Audit Intelligence” architecture built with synthetic data, featuring an orchestrated team of AI agents coordinated by an Audit Lead and visualized through an interactive chatbot interface. Attendees will gain insight into how specialized agents for Internal Audit, Regulatory Intelligence, SOP Compliance, and Quality Systems can collaborate to surface document interdependencies, link observations to source data, and proactively identify risks. The system leverages Retrieval-Augmented Generation (RAG) using both vector and graph databases, enhancing traceability, context, and document navigation for auditors. Importantly, this session also shows how AI components can be conceptually mapped to GAMP 5 and CSV controls—using URS-based design, guardrails, and synthetic validation artifacts. Whether you're in QA, Compliance, CSV, or AI governance, this talk provides a structured, low-risk pathway to explore and evaluate AI in Quality Systems—without compromising GxP expectations.

Operationalizing Artificial Intelligence in pharmaceutical manufacturing for Automated Visual Inspection represents a critical regulatory challenge, given that the 100% visual inspection of parenteral drug products is an important quality assurance step mandated by global pharmacopeias and GMP guidelines to ensure product quality and patient safety. AI, particularly machine learning, overcomes the key limitations of traditional inspection methods, such as high false ejection rates and an inability to adapt to new defects. By learning complex visual patterns, AI systems enable higher detection accuracy to boost production yield.

Bridging the gap from algorithm development to an audit-ready state requires implementing a robust operational framework rooted in global compliance expectations. Regulatory bodies emphasize a risk-based lifecycle approach to manage model influence and decision consequence commensurate with the potential impact on product quality and patient safety. 

This presentation details the journey to regulatory compliance by establishing rigorous data and model governance focusing on ensuring training data is fit for use to mitigate bias, incorporating Explainable AI (XAI) techniques to ensure transparency and support human oversight, and integrating change management for model control and maintenance throughout the system’s validated lifecycle.

Bottom Line: Operationalizing AI within this high-risk GMP application may provide a blueprint for controlled, compliant industry-wide AI adoption.

The rapid maturation of artificial intelligence (AI) presents new opportunities for enhancing pharmacovigilance (PV) activities, yet the adoption in GxP-related environments requires rigorous, transparent, and lifecycle-driven evaluation. This work proposes a GAMP®-aligned methodology for assessing LLMs used in two high-impact PV use cases, Adverse Event (AE) classification and Literature Review & Surveillance, balancing innovation with regulatory expectations for patient safety, product quality, and data integrity. This evaluation framework integrates GAMP® 5 Second Edition principles, emphasizing risk-based assurance, clarity of intended use, traceability of requirements, and proportionate controls across the AI system lifecycle. The methodology defines objective benchmark criteria, for example, accuracy, precision/recall, F1-Score, confusion matrix analysis, inter-rater agreement, robustness to data variability, model drift indicators, and latency. In addition, some domain-specific metrics for LLMs, such as hallucination rate, factual consistency, prompt sensitivity, and explainability indicators (e.g., SHAP, attention-based evidence tracing), are incorporated to support defensible validation.

The evolution of Vision AI has moved past vision inspection to a tool that operators use to assure proper execution of aseptic procedures and to detect potential vial breakage. In aseptic pharmaceutical manufacturing, deviations—like first-air disruption or visual glove inspection—can cascade into major compliance risks and costly batch losses. Vial breakage is common and results in serious quality issues, lost production, and lost capacity. Vision AI is being used to detect potential misalignments and incorrect aseptic technique prior to the event, thereby alerting the operator to intervene prior to the event. We will show a video of a digital twin of an RABS to demonstrate the use cases that has used in an actual manufacturing environment. The solution provides real-time alerts to operators and supervisors on anomalies in the production line.

This presentation will offer valuable insights into:

• Technology background with the differentiation between visual inspection and Vision AI
• Share video demonstrations of aseptic technique compliance use cases
• Human-machine collaboration in cleanrooms: how operators are responding to AI-generated alerts
• Early wins and lessons learned in integrating Vision AI
• Discuss expected benefits and challenges

AI is transforming regulated manufacturing by accelerating operations, sharpening insights, and strengthening monitoring capabilities. Yet, organizations face unique challenges: ensuring compliance with strict regulations, maintaining transparency, and managing risk without slowing innovation.

Catalyx will share practical strategies to harness AI in highly regulated environments. We will explore:

• Meeting 21 CFR Part 11 requirements in AI-enabled systems
• Leveraging AI to improve efficiency without creating new vulnerabilities
• Adoption strategies that balance performance gains with risk management

Attendees will gain a clear roadmap for aligning innovation with compliance. Whether you are evaluating AI for the first time or optimizing established systems, you will leave with actionable steps to reduce risk and move forward with confidence.

Designing equipment for aseptic manufacturing is often an iterative, time-intensive effort that relies heavily on expert judgment and manual validation activities. These constraints make it difficult to ensure that a design is both optimal and fully aligned with microbiological risk realities and the principles of a contamination control strategy. This presentation introduces a deterministic approach to equipment and process design that incorporates probabilistic modelling and the targeted use of AI to enhance aseptic process understanding. Using a case study from the development of the Groninger next-generation filling line, the session will demonstrate how key validation components—including in-process controls, environmental and process monitoring strategies, and aseptic process simulations—can be generated through deterministic process models, systematically evaluated, and managed within a centralized digital repository. By moving from fragmented, manual routines to a structured and model-driven design framework, organizations gain greater transparency into microbiological quality and sterility assurance. The outcome is a validation process that is faster, more consistent, and substantially better at predicting and mitigating aseptic risks.

Artificial Intelligence is rapidly becoming embedded in pharmaceutical manufacturing, quality control, and decision-making systems. With this adoption comes increased regulatory scrutiny. Health authorities expect AI-enabled systems to demonstrate the same rigor, transparency, and lifecycle governance as traditional computerized systems. This session will present a structured playbook for inspection readiness when deploying AI in GMP environments, supported by short case studies that illustrate how these concepts are applied. The interactive panel discussion will highlight how to anticipate regulatory focus areas, embed AI oversight into the Quality Management System, and prepare subject matter experts with the right language, metrics, and supporting evidence. Practical tools will be shared, including frontline binders, inspection-critical documents, and model transparency records that align with global guidance such as FDA, EMA, Annex 22, ICH Q9(R1), and ISPE GAMP. Attendees will gain insights into explaining various validation approaches, dataset governance, bias detection, human-in-the-loop oversight, and drift monitoring. Case examples will demonstrate how companies use these approaches to satisfy inspection expectations. Whether a company is just beginning to explore AI or already has mature systems in place, this session provides actionable steps to strengthen oversight, increase inspector confidence, and position AI adoption as a compliance advantage rather than a risk.

Digital twins are increasingly recognized as a transformative technology in biomanufacturing and a key enabler of automations and the vision of lights-out manufacturing. Achieving this vision demands modeling technologies that can efficiently capture and predict process dynamics with high fidelity, while remaining sufficiently interpretable for operations and regulators. In this work, we highlight hybrid modelling as the key enabler of bioprocess digital twins across the process lifecycle, from R&D to commercial manufacturing. Through the presentation of newly published works and industrial case studies, we highlight: (1) the ability of hybrid models to efficiently generate and capture process knowledge during development, (2) their application in enabling “self-driving” development, and (3) the deployment of their stored knowledge as part of a trusted manufacturing digital twin, used in offline simulation, as well as online predictive monitoring with human-in-the-loop or closed-loop control.

As life-science organizations accelerate AI adoption, a structured risk assessment framework is critical to ensure GxP compliance, product quality, and patient safety. This session presents a six-step methodology that integrates IMDRF’s risk classification guidance with GAMP® 5 lifecycle principles. Attendees will learn how to: define AI system scope and intended use; conduct initial risk categorization; perform granular risk analysis across data, algorithm, and process components; develop targeted mitigation controls; implement continuous monitoring and validation; and establish a dynamic risk management lifecycle. Real-world case studies from manufacturing process optimization and clinical trial support illustrate how this framework addresses model drift, algorithmic bias, and evolving regulatory expectations. Participants will gain practical tools and templates to replicate this approach in their own organizations, enabling scalable, risk-based AI adoption that aligns with FDA and international standards while maintaining audit readiness and inspection confidence.

This presentation will discuss examples from pharmaceutical manufacturing to illustrate how manufacturers may establish the credibility and trustworthiness of their artificial intelligence (AI) tools. Focusing on the use of AI to monitor production in real time, the presentation will discuss types of data and processes employed to address credibility, as well as the monitoring and validation steps used to promote trustworthiness. Examples may be drawn from process analytical technology, including near-infrared spectroscopy and digital twins. Analysis of the examples will contribute to greater understanding as to what information may help establish AI credibility and trustworthiness, as well as ways to evaluate this information and draw conclusions about AI tools.

GenAI has emerged as the primary instrument for Life Sciences organizations exploring and building AI solutions, yet its stochastic nature challenges established principles of computer system validation (CSV). Unlike deterministic, rule-based systems, LLMs produce variable outputs, making them difficult to verify, qualify, and control in alignment with GxP expectations. Some frameworks, such as the proposed Annex 22 guidance, significantly limit the use of LLMs in GxP applications. Today’s practices of controlling these types of models are rudimentary at best, mainly relying on the concept of human-in-the-loop. This session explores how the industry can and should improve and navigate the growing gap between LLM adoption and the regulatory demand for evidential control. Through practical examples, we will explore approaches to measuring LLMs, defining performance criteria, and establishing evaluation workflows. We will also discuss the limitations of these evaluation metrics and suggest appropriate controls for them. The session concludes with a discussion of the path forward — what a risk-based, evidence-driven framework for GenAI validation could look like, and how to bring and uphold scientific rigor to enable the widespread use of these technologies.

AI is rapidly reshaping the medical device landscape, creating new opportunities while also increasing regulatory complexity. This fireside chat will explore how expectations are evolving from a regulator’s perspective across AI in the QMS, AI enabled SaMD, and AI integrated into medical devices. Drawing on real world observations, we will highlight common pitfalls identified during assessments, including challenges in data management, model validation, cybersecurity, and lifecycle control. We will also discuss how EU MDR and the EU AI Act are converging to raise expectations for risk management, transparency, and post market surveillance. Attendees will gain practical insight into where manufacturers most often fall short and how to better prepare for evolving regulator expectations.