Responsible AI Summit Main Conference Day 2 - Tuesday 22 September

For an AI agent, even more than other systems, “Responsible AI" is indistinguishable from "working correctly. “As a world leader in responsible AI, Mastercard is advancing governance for the next era of autonomous systems. As AI evolves from predictive tools to agentic systems that act independently, governance must evolve with equal clarity and precision. This session discusses the additional risks agents bring above traditional and generative AI, including Mastercard’s practical approach to managing these at scale.

• The case for Responsible Agentic AI.

• Agentic Risks and Risk Taxonomy.

• Developing Trustworthy, Responsible and Safe AI Agents

Richard Boorman

Director of AI Governance
Mastercard

See Bio

AI literacy is the operating layer of responsible AI: shaping how teams interpret insights, exercise judgment, and maintain human oversight as AI becomes embedded in workflows. Moving beyond static adoption metrics, organisations must focus on cultural readiness and measurable capability shifts that influence governance and decision quality. This session explores practical approaches to literacy and culture so that AI deployment strengthens accountability and operational performance rather than simply increasing usage.

• Measuring literacy and decision impact beyond adoption metrics.

• Cultivating culture and shared understanding for responsible outcomes.

• Integrating literacy with operations to strengthen oversight and judgment.

James Fletcher

Responsible AI Lead
BBC

See Bio

Ger Janssen

AI Ethics & Compliance Lead
Philips

See Bio

The EU AI Act mandates human oversight, yet in practice, this often becomes a junior staff becoming an “AI checker” rubber-stamping automated outputs. In this candid session, Dr David Crelley, Head of Responsible AI & Data at Admiral Group, challenges the compliance-driven interpretation of human-in-the-loop and examines why oversight designed for efficiency frequently undermines effectiveness and accountability. He will share how his team is rethinking oversight as proactive engagement, using judge LLMs to do the hard lifting, deliberately introducing friction, cultural ownership, and meaningful intervention points into AI-enabled processes. David offers practical insight into how to design oversight models that create genuine human engagement rather than passive validation.

• Moving from passive validation to active engagement.

• Using LLMs to do the basic checks.

• Designing friction to strengthen human judgement.

• Embedding cultural ownership into AI oversight.

David Crelley

Head of Responsible AI and Data
Admiral Group

See Bio

Every organisation says data is strategic, yet most still run on brittle pipelines, stale datasets, and underfunded infrastructure. As automation, real time AI systems, and agentic workflows explode, the gap between what models need and what data teams receive is widening fast. In this panel, speakers will unpack what “automated data” truly requires today, how to quantify its ROI, and how to build foundations that make Responsible AI actually possible.

Andy Chi

Senior Legal Counsel and Data & AI Governance Lead
SHEIN

See Bio

Accuracy scores and leaderboard rankings only tell a small part of the story. In real environments, LLMs can be influenced, manipulated, and pushed into failure modes that standard evaluation simply doesn't capture. This session focuses on how LLMs behave once deployed, especially from a risk and security perspective. Drawing on practical examples such as prompt injection, RAG poisoning, and everyday failure patterns, we'll explore where current evaluation approaches fall short and what organisations should be testing instead. Rather than asking "is the model accurate?", the session reframes the question to: "how does the model behave under pressure, and where does it break?"

• Understanding how LLMs can be influenced and where real risks emerge.

• Identifying failure modes such as prompt injection, RAG poisoning, and hidden vulnerabilities.

• Rethinking evaluation to reflect real usage, adversarial conditions, and enterprise risk.

Shay Weiss

Head of Ireland’s Engineering, DevOps and Product, Director WBA Digital
Walgreens Boots Alliance

See Bio

As AI systems become more conversational, expressive, and agentic, the push to humanise them is accelerating. In China, widespread deployment of highly human-like AI across daily life has already triggered regulatory action aimed at curbing over-anthropomorphism and protecting users from misplaced trust. Europe now has a choice: learn early or regulate late. In this session, Sarah Mathews, Group Responsible AI Manager at Adecco Group, explores what responsible design looks like before AI systems blur the line between tool and perceived actor, and how transparency, human-centricity, and governance must evolve as agentic capabilities scale.

• Anticipating regulatory risk from humanised AI systems.

• Designing transparency into increasingly agentic experiences.

• Protecting human agency as AI embeds into daily life.

Sarah Mathews

Group Responsible AI Manager
The Adecco Group

See Bio

As AI increasingly shapes core product, operational and strategic decisions, many organisations still treat ethics as a policy artefact or compliance exercise - often detached from the reality of day to day decision-making. Roos Brekelmans, Digital Innovation Accelerator at Vattenfall, explores the ethical, strategic and trust risks created by this gap. Rather than framing ethical AI as a control or governance problem, she reframes it as a practical business discipline - and a conversation skill. Drawing on hands-on innovation experience and academic foundations in technology ethics, this session moves beyond abstract principles to focus on concrete, repeatable practices that teams and leaders can embed directly into their innovation pipelines. Using AI responsibly is positioned not as a brake on progress, but as a lever for resilient, scalable, and strategically sound innovation.

• Embedding ethics into everyday and strategic business decisions.

• Using values to navigate real AI trade-offs under uncertainty.

• Linking responsible innovation directly to enterprise value and trust at scale.

Roos Brekelmans

Digital Innovation Accelerator
Vattenfall

See Bio

As AI systems scale, the human labour powering them, data annotators, content moderators, and crowd workers, becomes operationally invisible. Yet laws such as the Corporate Sustainability Due Diligence Directive make oversight of supply chains a legal obligation, not a reputational choice. Ethical risk in AI no longer sits only within the model; it extends across global labour networks that train, filter, and sustain these systems.

This panel examines why auditing your AI supply chain matters, and what meaningful accountability looks like in practice. Speakers will highlight how standards-based evaluation, like Fairwork AI, can expose hidden labour risks, benchmark working conditions, and create enforceable transparency. The discussion moves beyond awareness to explore due diligence, procurement leverage, and how organisations can align AI deployment with fair work principles under emerging regulatory pressure.

• Exposing hidden labour across AI supply chains.

• Using certification frameworks to benchmark labour standards.

• Embedding due diligence into AI procurement decisions.

Mark Graham

Professor of Internet Geography, Director of Fairwork
University of Oxford, Fairwork

See Bio

Krisztina Pinter

Head of Global Service Planning and Data Governance
SONY

See Bio

Generative AI is accelerating across insurance, bringing new capabilities and new risks to underwriting, claims, and customer operations. In this evolving environment, model risk functions must ensure that governance frameworks, validation processes, and lifecycle controls can adapt without losing rigour.

Nadeem Chaudhry

Model Risk and AI Governance Lead
Aviva

See Bio

Roberto Martin-Reguera

Group Model Validation Director
Aviva

See Bio

As generative AI embeds across enterprise systems, bias becomes structural: shaping decisions, communications, and customer outcomes at scale. What begins as a model limitation can quickly become a systemic risk, reinforced by data provenance, prompt design, feedback loops, and organisational incentives. This panel takes a critical look at bias as a socio-technical challenge spanning the full AI lifecycle. Moving beyond surface-level fairness claims, panellists will examine how to measure structural and emergent bias in production, define meaningful accountability, and govern generative systems that continuously adapt and influence behaviour.

• Examining bias across the AI lifecycle.

• Measuring structural and emergent bias in production.

• Embedding accountability beyond technical fixes.

Olu Akinyede

Data Privacy, Data Governance and AI Ethics
Aviva

See Bio

Clara Higuera Cabañes

Responsible AI Program Lead
BBVA

See Bio

Uthman Ali

Associate Fellow
University of Oxford

See Bio