Share with your CISO
SailPoint is making a pointed argument that AI security isn’t a post-deployment audit item but a condition of scale itself. ASEAN MD Eric Kong cites SailPoint research showing 80% of surveyed organizations have already experienced data breaches or unintended data releases caused by AI agents, with more than 53% of those agents accessing sensitive company data daily. Machine identities now outnumber human identities by 40 to one in some enterprise environments, a ratio that grows every time a new agent gets provisioned.
What this means for your business
The 40-to-one machine-to-human identity ratio is the number that should recalibrate your threat model right now. If your identity and access management program was sized and structured for a workforce of people, it was never designed for the volume, speed, or autonomy of an agent fleet. Enterprises already running multi-cloud, multi-application environments are the ones most exposed, because each new agent introduced into that architecture becomes an independently addressable attack surface operating around the clock without fatigue.
The least-privilege and just-in-time access principles Kong recommends aren’t new, but applying them to AI agents is genuinely harder than applying them to humans. A human employee has a job title that maps roughly to a permission set. An AI agent has a task graph that can shift dynamically, meaning static role-based controls, the backbone of most enterprise identity programs, break down in exactly the moments when an agent is doing something novel or high-stakes. The 80% breach figure, sourced from SailPoint’s own research and therefore carrying the optimism of a vendor selling the cure, is still directionally credible given how few enterprises have formal agent governance programs in place.
The more telling signal here isn’t the breach statistic but the structural shift underneath it. When agents outnumber employees by 40 to one, the CISO’s job stops being primarily about authenticating people and starts being about governing autonomous processes. That’s a different discipline, a different toolset, and likely a different budget conversation. The falsification condition for Kong’s “identity-first AI strategy” argument is simple: if enterprises can demonstrate low breach rates with lightweight, task-scoped security controls rather than a unified identity platform, the case for a full SailPoint-style solution weakens considerably. Until that evidence exists, the architecture he’s describing is the right starting point.
Concept deep-dive: Just-in-time access
Just-in-time access means permissions are granted to a system or user only for the duration of a specific task, then automatically revoked when the task completes, rather than sitting open indefinitely. Think of it as a hotel keycard that expires at checkout instead of one that works forever. For AI agents, which can operate continuously and touch multiple sensitive systems in a single workflow, permanent standing access is a liability. Just-in-time access shrinks the window during which a compromised or misconfigured agent can cause damage.
Based on reporting from AI security seen as key to scaling enterprise adoption, originally published 2026-09-09 05:46:00.
