AI Agents Are Your Fastest-Growing Identity Risk

AI Agents Are Your Fastest-Growing Identity Risk

Every enterprise security team has spent the last decade hardening how humans log in: multi-factor authentication, privileged access reviews, SIEM-monitored behavior. According to the Cloud Security Alliance's May 2026 whitepaper on non-human identity governance, that investment addresses only the smaller half of the identity problem. Non-human identities — service accounts, API keys, OAuth tokens, and now the credentials AI agents hold — outnumber human users by an average of 45 to 1 across the enterprise, and by as much as 144 to 1 in cloud-native environments, up from a 92-to-1 ratio in the first half of 2024. That's a 56% jump in the ratio in a single year, and AI agents are the fastest-growing piece of it.

Machine Identities Have Quietly Become the Majority

The scale is easy to understate until you see a real audit. CSA's whitepaper cites a published case study of a Fortune 500 financial institution that found over 4.2 million non-human identities against roughly 50,000 human user accounts. Veeam's own governance guide puts a similar number on the broader trend, citing KPMG's Cybersecurity Considerations 2026 report: the non-human-to-human identity ratio sits above 80 to 1 in the average enterprise, and machine identities grew from roughly 50,000 to 250,000 per organization between 2021 and 2025.

AI Agents Aren't Like the Identities Before Them

Every wave of enterprise technology has created a new category of machine identity, and every generation has outpaced the governance built for the one before it — service accounts, then API keys and cloud IAM roles, then container workload identities. CSA's whitepaper argues AI agents are a genuinely different problem, not just a bigger version of the old one, because they're not static credential holders. An agent can encounter a task mid-execution that requires access it wasn't provisioned for, and depending on its architecture, it can request that access, assume a different role, or invoke a privileged API it wasn't expected to need at deployment time — meaning the blast radius of an agent's credential isn't fixed when the credential is issued, it expands as the agent operates. Multi-agent orchestration compounds this: a single workflow can spawn dozens of ephemeral sub-agents, each needing its own identity, with lifecycles measured in seconds rather than years. By May 2026, per the same whitepaper, Microsoft Copilot Studio users alone had collectively created more than one million AI agents, and Gartner projects 33% of enterprise applications will incorporate agentic AI by 2028, up from less than 1% in 2024.

Nobody Owns the Credentials, and Nobody's Watching Them

The governance data is the uncomfortable part. CSA's own 2024 survey found only 15% of organizations feel highly confident in their ability to prevent attacks based on non-human identities, and a separate 2026 CSA analysis of token sprawl found more than 16% of organizations don't track the creation of AI-related identities at all. The Cyber Strategy Institute's 2026 NHI Reality Report, cited in the same whitepaper, found 78% of organizations have no documented policy for creating or removing an AI agent's identity in the first place. The consequences show up in breach data: IBM's Cost of a Data Breach Report 2025 found 97% of organizations that suffered an AI-related security breach lacked proper AI access controls. And the exposure keeps growing at the credential layer — GitGuardian's State of Secrets Sprawl 2026 report found 28.65 million hardcoded secrets were added to public GitHub repositories in 2025 alone, a 34% year-over-year increase and the largest single-year jump on record. Within that figure, AI-related secrets — keys for AI APIs, agent configuration tokens, LLM service credentials — accounted for more than 1.27 million exposures, an 81% year-over-year increase and the fastest-growing category of any credential type tracked.

What Governance Actually Looks Like

Neither report treats this as unsolvable. Veeam's guide lays out six practices that show up consistently across organizations that have gotten ahead of the problem: build a continuous inventory of every non-human actor (service accounts, workloads, bots, and agents alike), map exactly which credentials each one uses, assign a human owner to every single identity, apply least-privilege scoping rather than the broad access that's fastest to set up, rotate and expire credentials on a schedule instead of letting them persist indefinitely, and extend Zero Trust verification — checking identity on every request rather than trusting one upstream login — to machine identities the same way it's applied to human ones. None of these are exotic; they're the same discipline security teams already apply to people, just not yet applied consistently to the identities that now outnumber people by 45 to 1.

What This Means for Teams Deploying Agents Going Forward

The organizations that CSA and Veeam describe as exposed aren't the ones moving slowly on AI agents — they're the ones moving fast without first deciding who owns each agent's access, how long it should last, and who's checking it. Given that machine identities already outnumber human ones by a wide margin and AI agents are the single fastest-growing category of them, an agent workforce without a credential owner, a scoped permission set, and a documented lifecycle isn't a minor gap to clean up later. It's the same governance question every prior wave of enterprise technology eventually forced — service accounts, API keys, cloud IAM — just faster and higher-stakes this time.

FAQ

Are AI agents really a bigger security risk than other machine identities? They're a different kind of risk, not just a bigger one. Traditional service accounts and API keys are static — the same permissions every time. AI agents are non-deterministic and can acquire additional access dynamically mid-task, spawn sub-agents with their own credentials, and chain actions across systems in ways that weren't fully defined when the agent was deployed, per the Cloud Security Alliance's May 2026 whitepaper on the topic.

How many machine identities does a typical company actually have? More than most security teams can currently count. KPMG's 2026 Cybersecurity Considerations report puts the average enterprise ratio above 80 non-human identities per human, and machine identity counts per organization grew roughly fivefold, from about 50,000 to 250,000, between 2021 and 2025.

What's the actual fix, in practice? Treat every AI agent's credential the way you'd treat a new employee's access: inventory it, assign a human owner, scope it to least privilege, set it to expire and rotate, and monitor its activity — the same governance discipline already applied to human accounts, extended to the identities that now vastly outnumber them.