Skip to content 
Every AI agent an employee connects to Slack, Salesforce, or an internal database creates a new credential — an OAuth token, an API key, a service account. Almost none of them get added to a central identity inventory. Multiply that across a workforce experimenting with agents on its own initiative, and enterprises are accumulating a population of machine identities that security teams didn’t provision, don’t track, and often don’t know exists.
The Scale of the Problem
Non-human identities (NHIs) — service accounts, API keys, bots, and now autonomous AI agents — already outnumber human identities in most enterprises, and the gap is widening fast. Estimates vary by source, but they all point in the same direction: NHIs outnumber human accounts anywhere from 45:1 to over 100:1 in modern enterprise environments, and one analysis puts the ratio as high as 82 machine identities for every human one.
The growth rate is what makes this a 2026 story rather than a background IAM concern. A SANS Institute survey of over 500 security professionals found that 76% of organizations report growth in non-human identities, with 74% already running AI agents or automations that require their own credentials — a shift that has caused NHI populations to double or triple inside many organizations in a short window. Separately, the 2026 Verizon DBIR found that employee use of unapproved (“shadow”) AI tools tripled to affect 45% of the workforce, and each of those unsanctioned deployments tends to generate credentials stored outside normal secrets management, never inventoried and never rotated.
Why this is a governance gap, not just a scale problem
Most organizations agree the risk is real. Fewer have done anything about it:
- 92% of organizations say governing AI agents is critical, but only 44% have implemented any policy to do so, according to Omada Identity’s State of Identity Governance research.
- 92% also say their existing IAM tooling cannot manage AI agent identities, per Security Boulevard’s analysis of the agent identity problem — the tools built for human logins don’t map cleanly onto machine-to-machine access.
- Only 5.7% of organizations report full visibility into their service accounts, and a Cloud Security Alliance survey found 68% of organizations cannot reliably distinguish AI agent activity from human activity in their logs.
- 16% of organizations don’t track the creation of AI-related identities at all, per a 2026 CSA token-sprawl analysis — meaning the credential doesn’t just go unmanaged, it goes unseen.
That combination — rapid credential growth, low visibility, and tooling built for a different problem — is what makes NHI governance, in the Cloud Security Alliance’s framing, the defining security gap of the agentic AI era.
Where the actual damage happens
The risk isn’t theoretical. A few patterns show up repeatedly in 2026 incident data:
Over-permissioned agents. A Non-Human Identity Management Group report found that 73% of secrets held by NHIs carry excessive permissions, and over 5.5% of AWS machine identities hold full administrative privileges. When an AI agent inherits a credential at that level, the line between “authorized to do its job” and “authorized to do anything” effectively disappears — a pattern OWASP’s Agentic Security Initiative now classifies as ASI03, Identity and Privilege Abuse.
Credentials that never get rotated. Attackers reportedly exploit exposed AWS credentials within an average of 17 minutes, while nearly a quarter of organizations take more than 24 hours to rotate them once exposure is discovered. Separately, 64% of secrets confirmed exposed in 2022 were reportedly still valid four years later.
Unauthenticated integration points. A study of nearly 8,000 live Model Context Protocol (MCP) servers found that 40% had zero authentication configured, and every OAuth-enabled server tested carried at least one flaw. Real incidents have followed this exact pattern — a compromised third-party OAuth integration, not malware, exposing database secrets and signing keys.
Concentrated blast radius. One 2026 identity report found that a mere 0.01% of non-human identities control 80% of cloud resources — meaning a small number of poorly governed machine credentials can carry an outsized share of enterprise risk.
Building an NHI governance program
Security teams that are ahead of this problem are treating agent identity as a first-class IAM discipline, not an extension of secrets management. Five moves show up consistently in current guidance:
- Inventory before you police. You cannot govern what you cannot see — start by discovering every API key, service account, and agent credential currently in use, including the ones created outside formal channels.
- Assign an owner and a purpose to every agent identity, the same way a privileged human account requires one. An NHI without a named owner and a defined scope is a standing liability.
- Scope permissions to task, not convenience. Default-to-admin credentials are the single biggest amplifier of NHI risk; least-privilege scoping should apply as strictly to agents as to people.
- Rotate and expire credentials on a schedule, not on discovery. Treat non-rotation as an active vulnerability rather than a hygiene nice-to-have.
- Extend detection to distinguish agent behavior from human behavior. If your SIEM can’t tell the difference between an employee and an agent acting under that employee’s integration, you can’t detect misuse either.
Identity providers are starting to respond directly: Auth0 shipped “Auth for MCP” in mid-2026, and Okta released an MCP server that lets agents interact with Okta’s management APIs under scoped, least-privilege access enforced at each tool call. These are useful authentication building blocks — but as several 2026 analyses note, authentication alone doesn’t govern the full identity lifecycle. Discovery, ownership, rotation, and offboarding still have to be built on top.
What security leaders should prioritize now?
CSO Online named NHI governance one of the top agenda items at RSAC 2026 — the first time the topic has been addressed at that scale on the conference’s main stage, which is a reasonable proxy for how quickly this has moved from niche to board-level. For CISOs and IAM leaders, the practical starting point is:
- Run a full NHI discovery pass, including shadow AI tools employees have connected without formal approval.
- Bring agent identities into the same Zero Trust-style continuous verification model already being applied to human SSO — least privilege, short-lived credentials, and behavior-based risk scoring.
- Build offboarding and credential-expiry into agent lifecycle management from day one, rather than retrofitting it after the first incident.
Human identity governance took two decades of MFA mandates, PAM rollouts, and SIEM tuning to mature. Non-human identity is compressing that same curve into a couple of years — and right now, most organizations are governing a small fraction of the identities actually operating inside their environment.