TELUS is Canada's leading telecom provider and a global leader in contact center BPO, content moderation, digital transformation, and AI model training. In 2025 alone, TELUS generated more than two trillion AI tokens across 100+ production use cases. That scale exposed a hard truth: the vulnerabilities hiding inside AI systems are far more common, and far more dangerous, than most security teams realize. So why does AI security testing still miss them, and where are the blind spots?

This session breaks down where manual and automated testing fall short, including why off-the-shelf and fine-tuned models produce weak attack simulations that leave real exposure undetected. We'll show how these gaps widen when moving from generative AI applications that deliver information to AI agents that take action, and why testing built for yesterday's risk can't keep pace with today's. It's grounded in current academic research and real-world testing of large-scale AI production systems.

CISOs and AppSec leaders will leave able to spot vulnerabilities faster, evaluate testing solutions more critically, and manage exposure introduced by the latest AI capabilities before they become the next hack or headline.