At the moment’s episode is a dialog with Bret Kinsella, recorded whereas he was in Las Vegas for CES and making ready to step onto the AI stage. Bret brings a uncommon mixture of long-term perspective and hands-on expertise. As Basic Supervisor of Gas iX at TELUS Digitalhe operates generative AI methods at a scale most enterprises by no means see, processing trillions of tokens and delivering measurable enterprise outcomes for world organizations. That vantage level offers him a transparent view of each the promise of generative AI and the uncomfortable truths many groups are nonetheless avoiding.
Collectively, we unpack why generative AI breaks so lots of the assumptions safety groups have relied on for many years. Bret explains why these methods are probabilistic reasonably than deterministic, and the way that single shift creates what he calls an unbounded assault floor.
Customers are now not restricted to predefined buttons or workflows, and outputs are now not constrained to a hard and fast database. The identical immediate can succeed or fail relying on delicate modifications, which makes single-pass testing and checkbox compliance dangerously deceptive. When you’ve got ever puzzled why an AI system feels protected sooner or later and unpredictable the following, this dialog gives a grounded rationalization.
We additionally discover why specializing in the mannequin alone misses the true danger. Bret makes a powerful case that the mannequin is just one a part of a a lot bigger system formed by system prompts, linked information sources, instruments, and guardrails. Change any a kind of parts and habits shifts. That is why automated, steady purple teaming has turn out to be unavoidable.
Bret shares how Telus Digital’s Fortify AI assault mannequin uncovered lots of of vulnerabilities in hours, far past what human groups may realistically floor on their very own. But automation shouldn’t be the tip of the story. The ultimate choices nonetheless depend upon individuals who perceive context, trade-offs, and enterprise affect.
All through the dialogue, we return to a easy however uncomfortable concept. AI security shouldn’t be one thing you bolt on after deployment. It calls for a unique mindset, broader testing, repeated validation, and ongoing human judgment. For leaders shifting from experimentation to real-world deployment, this episode is a clear-eyed take a look at what accountable progress really requires.
As extra organizations rush to deploy brokers and autonomous methods in 2026, are we really ready for software program that learns, adapts, and infrequently surprises us? What does that imply for a way you check and belief AI inside your individual enterprise?
Helpful Hyperlinks
Subscribe to the Tech Talks Each day Podcast
