Every tech-industry appreciation day comes with a press release, and AI Appreciation Day is no exception. But the note from cybersecurity firm Illumio reads less like a celebration than a warning: the same AI that companies are racing to deploy is also handing attackers a faster way in.
The argument is straightforward. Frontier AI is automating the grunt work of an attack — reconnaissance, vulnerability discovery, the probing that once demanded time and real expertise — and compressing it into minutes. At the same time, the scramble to bolt AI onto everything is surfacing weaknesses that were already there: patchy visibility into where sensitive data lives, sloppy identity management, and sprawling system complexity that nobody fully maps.
“AI is transforming cyber defense, but it is also dramatically expanding the attack surface,” said Michael Adjei, Director of Systems Engineering at Illumio. “It has democratised knowledge and capability, bringing speed and power to both attackers and defenders. Power, however, is nothing without control.”
Assume the breach
As AI systems reach across interconnected cloud, on-premises and hybrid environments, Illumio argues, the gap between a minor misconfiguration and a full-blown incident keeps narrowing. Its central pitch is that prevention alone is no longer enough. Organisations should assume faster attacks will eventually slip past detection, and plan for containment and impact rather than betting everything on keeping intruders out.
That framing is, conveniently, Illumio’s entire business — the company sells microsegmentation designed to stop attackers from moving laterally once they are inside. But the underlying point is hard to dispute, and it is one plenty of independent security researchers have made too: “AI-ready” is too often treated as a tooling problem when it is really a fundamentals problem.
To that end, the firm’s checklist is less about buying new products than fixing old habits. It recommends knowing where critical data sits and how it moves, mapping the APIs and third-party integrations that AI services depend on, tightening identity and credential controls across environments, and promptly revoking access when people leave or systems are retired. Where something cannot be patched quickly, it suggests restricting access to it; and where an account is compromised, segmentation should stop the blast radius from spreading across the network.
None of this is new advice. What has changed is the clock. With attackers now able to weaponise the same models defenders are excited about, the margin for leaving the basics half-finished is shrinking — and a marketing holiday is as good a reminder as any.
