Agentic AI Forces Zero Trust From Long-Term Goal to Immediate Requirement
The Development
Ping Identity CEO Andre Durand has made the case that agentic AI has fundamentally broken the assumptions underlying traditional identity and access management. Where a compromised human account might generate harmful actions over minutes or hours, an AI agent can execute a thousand discrete actions in five minutes — each one potentially carrying its own access approval. The compounding risk emerges from how enterprises currently provision agents: shared service accounts, cloned human logins, and API keys embedded in source code create standing permissions that agents can hold indefinitely. Durand identifies API gateways and agent gateways in front of MCP servers as the enforcement choke points where real-time policy can be applied, and calls for every agent to carry its own distinct identity rather than inheriting credentials from a human user.
Our Take
The deeper problem Durand is naming is not a security configuration issue — it is an architectural one. Every enterprise that has been running AI agents under legacy IAM assumptions has been quietly accumulating a permission surface nobody audited, because the approvals looked routine in isolation. The practical enforcement framework he describes — distinct agent identities, per-action authorization at API and MCP gateways, independent multi-agent review structures — is not theoretical. These are deployable controls. The window to build this correctly before agentic adoption scales into the thousands of agents is closing fast. Retrofitting identity architecture into a mature agentic deployment will be significantly more expensive than building it in now, and the blast radius of getting it wrong scales with the number of agents, not the number of humans.
What Changed
Enterprises can now enforce authorization at the level of individual agent actions — not sessions — by applying real-time risk and fraud signals at API gateways and MCP server gateways. This moves identity control from a login-time decision to a per-action decision, closing the standing-permission window that agentic workflows currently exploit.
Marketing Impact
Marketing operations teams running AI agents across CRM, content workflows, and campaign automation are accumulating unaudited access permissions at scale. Any agent operating under a shared service account or embedded API key represents an open standing permission that current martech security reviews are not designed to catch.
Competitive Implication
Enterprises that build agent-native identity infrastructure now — distinct agent identities, per-action authorization, gateway enforcement — will be able to scale agentic deployments faster and with lower regulatory exposure than competitors still operating on session-based IAM. Those continuing with legacy provisioning face compounding retrofit costs and growing audit liability as agentic volumes increase.
Strategic Outlook
Identity vendors with agent-native architectures — Ping Identity among them — are positioned to take share from legacy IAM providers as enterprises begin formal agentic security audits in Q4 2026 and into 2027. Expect enterprise procurement cycles to add agent identity requirements to IAM RFPs within the next two quarters as internal security teams flag permission sprawl.
The Exploit
Action Item
CISOs and marketing technology leads should jointly audit every AI agent currently operating under a shared service account or embedded API key, then prioritize migration to individual agent identities before Q4 2026 budget cycles lock infrastructure spend.