Key Takeaways
Look at the past couple of years, and it’s obvious that everybody is eager to check off the AI box — from AI builders to AI copilots to AI site generation.
Everything. AI.
But there’s a serious misconception here, says Jay Bavisi, the founder of EC-Council, a cybersecurity training program: Many folks believe that cloud security expertise can translate to AI security.
On the surface, it actually kind of makes sense. AI runs in the cloud, using all the familiar pieces, like compute, storage, and APIs. Naturally, people began assuming that as long as you’ve secured the environment — the cloud — you’ve also secured the system — AI.
“Most organizations don’t realize they have a problem until they’re already halfway through an AI deployment and suddenly asking questions their infrastructure teams can’t answer,” Bavisi told us.
If this feels familiar, it’s because it is. We went through the same thing in the early days of the cloud, when everyone assumed that if the infrastructure was secure, the system would be, too.
That, of course, wasn’t the case. The issue is that as humans, we’re more likely to learn from our mistakes instead of always having foresight. Which means we actually have to make those mistakes first.
After all, airports didn’t always scan luggage; multifactor authentication didn’t exist until passwords alone weren’t making the cut anymore.
AI is, naturally, following the same pattern. But unlike the workloads we know and love, AI systems introduce a myriad of new risk layers, like model manipulation and data poisoning in LLMs. Most providers just aren’t prepared.
It’s why Bavisi is so vocal in warning about the point when you realize the mistake is already too big to fix.
“Right now, most providers are offering customers Band-Aid solutions when they should be rearchitecting their platforms from the ground up,” he told us.
AI Has a Mind Of Its Own
AI essentially has a mind of its own. It’s like dropping a new kid into a classroom and hoping they just fit in. Except this one doesn’t follow school rules, and no one really knows how they’ll react until they’re already in the room.
Educated teams, Bavisi said, know that cloud security expertise doesn’t transfer to AI workloads.
Instead, they’re “looking for frameworks that actually address the full lifecycle rather than just the infrastructure layer.”
Take ChatGPT hallucinations, for example. You might ask it a simple question and it gives you an answer that sounds totally believable. Maybe it invents a statistic, misquotes a source, or confidently explains something that never even happened.
At the surface-level, nothing is actually wrong: The system is working as it should. But since it’s hallucinating, users began asking where these answers came from — there was no simple trail to show how or why these models invented their own answer.
I don’t know about you, but I would blame whoever sold me the product.
It’s as Bavisi warns: “Vendors still treating AI as just another workload will discover too late that they’ve lost deals to competitors who recognized that AI governance isn’t optional; it’s the new basis of trust.”
But there’s still room for hosting providers to get this right — even if they’re not the ones building the AI, just putting it into production.
“The hosting providers who are winning right now aren’t the ones making bigger promises about AI support,” Bavisi said. “They’re the ones being radically honest about what they can and can’t control.”
Bavisi says this is what EC-Council is training its students to protect against, by addressing it with its ADG Framework:
- Adopt: Always test AI before you put it in your stack because it’s nearly impossible to back out once it’s in.
- Defend: Know how these systems can break because something will go wrong.
- Govern: Keep a handle on it well after launch; track what it’s doing, stay compliant, and don’t let it out of your sight just because it’s working (so far).
Because at the end of the day, customers might tolerate risk. In many instances, it’s par for the course to experience some risk. But ask a regulator or insurer and they’ll laugh in your face.
“Insurance and legal requirements will force vendor evaluation around AI risk faster than customer preference alone ever could,” Bavisi said. “Providers who acknowledge this gap and invest in redesigning their platforms for AI visibility will own the market.”
It’s always interesting to see how hosting providers are pulled into the security stack thanks to AI. The question is how they will continue to handle it.
