UNETSYSTEMS Jaeho Jang Senior Researcher

"As AI advances, so does the productivity of the people who use it. But unlike earlier waves of technology?which mainly drove ""vertical"" gains in speed and capability?AI now extends productivity ""horizontally"" as well, widening each persons reach. That horizontal expansion, however, translates directly into a larger organizational attack surface: both the points where confidential information can leak and the connection paths that attackers can exploit multiply at once. In particular, as employees use of shadow AI and the adoption of AI agents cause ""non-human identities"" to proliferate, the focus of defense is shifting from human logins alone to the interactions between machines and AI.
This talk divides that shift into two fronts, internal and external, and covers how to build guardrails and continuously validate that they actually hold.
On the internal front, we lay out how to rebuild the management of confidential data and identities around identity-centric Zero Trust. Viewing both people and AI agents as ""digital employees"" that are granted permissions to do their work, we approach it from two angles: an HR dimension?least-privilege and task-scoped credentials, plus onboarding and offboarding?and a network dimension that governs access and segmentation.
On the external front, we propose two forms of continuous validation. The first confirms whether the security controls you have already deployed actually work in practice. The second goes beyond the ""discoverable"" vulnerabilities that scanners list, to uncover the ones that are genuinely exploitable.
To treat AI not as something to be banned but as a tool for safely accelerating technological and business growth, we share a practical perspective on defending the inside and the outside?together, and continuously.
"