Jul 3, 2026

What it looks like when AI security works

AI vs AI
| Part
7

Reading time:

3 minutes

Tuesday morning. The phone didn't ring. The customer didn't leave. The Reddit thread didn't get written. That's what good looks like.

Read the full AI Security Series

1. The gap nobody's watching
93% confident. 29% prepared. One person jailbroke a chatbot and emptied ten government agencies. The distance between the story the boardroom hears and the one the incident report tells.

2. They're faster than you
27-second breakout times. AI-generated phishing that knows your team by name. The dark hoodie hacker lost their job to AI too. What replaced them doesn't sleep.

3. Your AI is their way in
76% of organizations have shadow AI. The threat isn't at the gate. It has a badge you issued. Vendor-embedded models, compromised plugins, and the tools your team invited in.

4. Sovereignty is an architecture decision
Everyone's saying sovereign. This article is about what happens when your customer asks and your answer needs caveats. Interchangeability, vendor independence, and what the CLOUD Act means for Canadian data.

5. The people problem
The AI security job posting asks for four careers in one person. That person doesn't exist. The work does. Hour banks, returnable context, and a PMO that holds it all together.

6. What to ask your vendor
Seven questions. Print them out. Bring them to the meeting. Look for straight answers, not carefully worded topic changes. Use them on everyone. Including us.

7. What it looks like when AI security works
Tuesday morning. The phone didn't ring. The customer didn't leave. The Reddit thread didn't get written. That's what good looks like.

Author

Dimitri Phalen is the marketing lead at ISM who prefers plain language over big claims. For years, he’s worked behind the scenes, translating messy, complex IT problems into something teams can actually use. If something sounds like it was written by someone with not enough coffee, who’s been sitting too close to the delivery team for too long, that’s probably his fault.

Tuesday morning. Early March. Still dark when the first person logs in from home, coffee in hand, cat doing something regrettable to the curtains. The overnight batch jobs finished. The AI inference pipeline processed its queue. The compliance dashboard updated itself. The security operations centre in Barrie flagged nothing unusual in the AI telemetry.

The CISO’s phone sat on the nightstand, quiet, so very boring.

A phishing campaign hit the building overnight. Personalized. One email per employee, built from scraped LinkedIn profiles and public project data, timed to arrive during the gap between the overnight shift and the morning. Every email was different. Every one looked real. Not one of them worked. The policy layer caught each compromised credential at the point of attempt, before a session could start, before a permission could be tested. The emails are in a quarantine log that somebody will review after standup. Nobody’s in a hurry. The infrastructure underneath held because it was built to hold, not because someone got lucky on a Tuesday.

Gary is using an AI tool right now, as we speak, to do something that would have taken him three days through the old process. The difference is: the tool was vetted. The data it can access is constrained by policy. The outputs are logged. When Gary inevitably finds a new tool on a Linkedin post and signs up with his work email, the governance layer will catch it, assess it, and either bring it into the fold or explain to Gary, gently, why this one doesn’t meet the bar. The cataloguing and classification work that made those guardrails possible was done months ago by people who’ve already moved on to the next engagement. Gary doesn’t feel restricted. Gary feels fast.

Gary also asked question seven in the vendor meeting last month. Where inference runs. What happens if the trade situation shifts. Someone noticed how much this mattered eventually. Gary’s team is having a good year.

The customer never posted on Reddit. There was nothing to post about. Their data is in Canada, managed by Canadians, under Canadian law. When a competitor’s data practices made the local news last month, a few customers checked in. The answer was one sentence. One sentence was all it took. They stayed. Not because they were locked in. Because they were looked after.

The board meeting next week has “AI expansion” on the agenda instead of “AI risk review.” The security team is planning, not reacting. The budget is stable because the architecture is stable. The team that built the AI security program didn’t come from a single unicorn hire who took eight months to find. It came from four specialists over four months, each one arriving with context, each one leaving without taking the institutional memory, a PMO holding it together, and an hour bank that meant the work started in week one instead of month nine.

ISM has been doing foundation work in Canadian enterprises for over fifty years. The name changes. Data governance. Cloud readiness. AI readiness. AI security. The work underneath is the same. It’s always been the same. Kyndryl brings the global depth, the alliances, the platform that’s been pressure-tested across thousands of environments. ISM turns that dug deep into something that works here. Local architecture. Stable technical ownership. Cleared personnel. People who answer the phone on a Saturday morning in February because they live here and the parking lot is already buriedin snow and they’re coming in anyway.

The prairie wind is doing something unfortunate to that parking lot right now, but inside, everything held.

93% confident. 29% prepared. We opened this series with those numbers because the gap between them is where everything breaks. Seven articles later, the gap closed. Not with a product launch. Not with a slide deck. With work. The kind that doesn’t get a keynote or a case study or a headline. The kind that produces a Tuesday morning where the phone didn’t ring, the customer didn’t leave, and the Reddit thread didn’t get written.

That’s what good looks like.

If you’ve read all seven, thank you. If you started here, the other six are worth your time. If you want to talk about what the work looks like in your environment, we’re not hard to find.

Cites

Kyndryl Security and Networks Readiness Report 2025-2026: 93% confidence, 29% preparedness

Kyndryl Bridge platform: AI-driven insights and operational automation at scale

Kyndryl Security Operations Centre, Barrie, Ontario: Canadian-staffed, Microsoft Azure-integrated

Want to understand how we work?

Every conversation starts with your environment. Not our product list.

Get in touch
ArrowArrow

It's not about the technology, or the processes. It's about your clients, your users, and the hard won trust you've built. Every team here keeps that at heart.

Mobile CTA