Responsible AI Summit North America: Day 2 - Wednesday, June 24, 2026

In highly regulated industries, the hardest part of AI is not building models - it’s proving you’re in control. Scotiabank is exploring how to measure and report AI risk in a way that stands up to regulators, especially as agentic systems and third-party AI services accelerate. This session unpacks the practical challenges of balancing mandatory compliance, vendor accountability, and speed-to-market, while building organisational trust, resilience, and security marked as “responsible”.

Aneta Osmola

Vice President Data and AI Risk
Scotiabank

See Bio

AI agents blur the boundary between technology and workforce: they operate within rules, yet exercise discretion in execution. This panel debates whether agentic systems should be governed like software, like employees, or through a new hybrid accountability model. Panellists examine how organizations define risk ownership and prioritize privacy when outcomes are unpredictable and failures carry operational, reputational, and regulatory consequences.

Hellena Crompton

Data Protection Officer UK&I
dentsu

See Bio

Cindy Tu

Director of IT & Data Audit
Capital One

See Bio

The rapid evolution of agentic AI, from single LLM-powered agents to coordinated crews and massive swarms, promises transformative autonomy in domains like supply chain, healthcare, and finance. Frameworks like CrewAI, LangGraph, and Google's ADK, combined with the Model Context Protocol (MCP) standard, enable dynamic agent-to-agent (A2A) collaboration and tool access. Yet this interconnected ecosystem introduces severe cascading risks: a single compromised agent can poison swarms via unsecured MCP calls, leading to data exfiltration, unauthorized actions, or ethical failures.

Veer Yedlapalli

Director of Product Security, Security Engineering and AI Security
Grainger

See Bio

Scaling AI across a workforce introduces real challenges around accountability, trust, and responsible use. In this session, Mark Lesiw, AI Enablement Program Owner at Xcel Energy, shares how Xcel Energy increased active enterprise AI usage from 33% to over 90% across 27,000 employees in under 12 months. Mark introduces Return on Employee, a term he uses to define AI value across a workforce, focusing on productivity, confidence, and decision quality, alongside traditional Return on Investment. He explains how Responsible AI was embedded from the start through literacy, tool-level guardrails, and human-in-the-loop design so that AI use could scale safely and deliver lasting business value.

Mark Lesiw

AI Enablement Program Owner
Xcel Energy

See Bio

Generative AI is advancing from drafting alerts to actively supporting cyber defence operations. This session explores real-world applications of threat detection, cyber hygiene, attack path analysis, and orchestrated response. It highlights practical insights into how organizations are using AI to responsibly enhance detection and strengthen cyber resilience.

Henry Awere

Responsible AI Operations Lead and Technology Risk
Canada Life

See Bio

Ayaz Minhas

AI Policy Manager
Meta

See Bio

Deploying AI in humanitarian and crisis settings requires systems that adapt to local realities without compromising responsibility. In this presentation, André Heller shares how the International Rescue Committee embedded aprendIA, an AI-powered chatbot, to support educators in Nigeria while navigating stakeholder needs, data protection laws, and resource constraints. The session explores how global responsible AI principles can be translated into practical, locally grounded deployments that scale safely across regions.

• Adapting AI systems to local cultural, regulatory, and operational contexts.

• Scaling AI solutions efficiently in constrained-resource environments.

• Applying global responsible AI guardrails to real-world humanitarian deployments.

André Heller Pérache

Director of AI
International Rescue Committee

See Bio

AI systems are now deeply embedded in core products, workflows, and decision-making, making governance a daily operating requirement, not a periodic review exercise. This panel examines how organizations are building cross-functional AI governance councils that are technically relevant, fast enough to keep pace with delivery teams, and clear on who has a seat and why. Speakers will break down how roles are defined in practice, how councils operate beyond high-risk use cases, and how governance models are built without forcing legacy policies onto AI systems that behave fundamentally differently.

Teresa Reilly

VP, Data + AI Governance
Kenvue

See Bio

Gary Brown

Chief Privacy Officer and AI Governance Corporate Lead
Westinghouse

See Bio

As agents and GenAI systems gain autonomy, traditional testing no longer exposes the highest-risk failures. This presentation explores adversarial testing techniques used to surface jailbreaks, tool misuse, data leakage, and unsafe autonomous behaviour. It focuses on how technical teams operationalise red teaming as a continuous capability rather than a one-time exercise.

Shone Mousseiri

Head of AI Model Validation and Governance
Manulife

See Bio

Unlike traditional software, agentic AI systems do not just execute instructions: they reason, act, and adapt through persistent interaction with tools and APIs. In this session, the speaker examines how this shift fundamentally breaks long-standing cybersecurity assumptions, particularly the reliance on prompt-based guardrails to control LLMs against attacks. They explore why guardrails are structurally insufficient against adaptive adversaries, and what it realistically means to defend systems where autonomy itself expands the attack surface.

• How agent infrastructure expands the cybersecurity attack surface.

• Why finite guardrails fail against adaptive prompt injection.

• Reframing AI security in the age of agentic.

Apostol Vassilev

Expert in Trustworthy and Responsible AI and Cybersecurity
NIST

See Bio