Agentic AI Security Risks: APAC Orgs Face Data Threats

Agentic AI, while boosting productivity, introduces new data security risks for organizations. These AI agents can accidentally or maliciously expose sensitive data due to overprivileged access. Here's the thing: a hidden instruction in an email could trick an AI assistant into exfiltrating company data. There are also risks from "Shadow AI," tools used without IT oversight, which can be unsafe and non-compliant. Jennifer Cheng, Director of Cybersecurity Strategy at Proofpoint, explains that the Agentic workspace involves people and AI agents working together across various platforms. Unlike earlier generative AI, agentic AI can reason, make decisions, and act autonomously. What's interesting is that every AI agent has an identity and access rights. If misconfigured or compromised, it can become a fast pathway for data exposure or credential abuse. Proofpoint's 2026 AI and Human Risk Report shows that about nine out of 10 organizations are deploying AI assistants beyond pilot programs, but about half are not confident their controls can detect compromised AI. Frontier AI models intensify these challenges, making it harder to mitigate compromised AI at scale. An employee connecting an AI scheduling agent, if compromised, could lead to attackers reading sensitive messages or impersonating users. The bottom line: organizations need to understand what AI can access to manage these new security risks effectively.