AI access is outpacing governance, and breach rates are rising

At a glance

What this is: NHI Mgmt Group's analysis, drawing on Cakewalk's research: AI-related access expansion is happening faster than governance, revocation, and shutdown processes can keep up in two 2026 surveys.

Why it matters: IAM, NHI, and AI governance teams need to treat AI access as an operational control problem, because broad standing access and slow revocation turn adoption into exposure.

By the numbers:

• Among organizations where AI significantly expanded the number of identities requiring access, breach rates reached 43% over the past twelve months.
• Only 19% of organizations say they fully govern their non-human identities.
• 76% cannot immediately revoke standing access once it is no longer needed.
• More than half of the professionals surveyed, 56%, did not know how long it would take to halt an AI system during a security incident.

Context

AI access governance is the set of controls that determine who or what can act inside enterprise systems, for how long, and with what revocation path. In this article, Cakewalk ties together two 2026 surveys to show that organizations are granting AI and agent access faster than they can govern or take it back.

The problem is not simple adoption. It is the combination of standing access, incomplete NHI governance, and weak shutdown readiness once AI systems are live. That creates an identity control gap that affects service accounts, agent credentials, and the broader governance model around autonomous work.

Key questions

Q: What breaks when AI access expands faster than governance can revoke it?

A: Standing access becomes persistent exposure instead of task-scoped permission. When revocation lags, overprivileged service accounts and agent credentials can keep acting long after their business need has ended, which widens blast radius and makes breach response slower and less certain.

Q: Why does broad AI access create risk for software and security teams?

A: Broad AI access increases risk because these systems can act on context, chain actions, and produce outputs that look credible even when they are wrong. If they reach beyond the intended boundary, they can expose information, introduce insecure decisions, and accelerate defect creation. The core issue is not AI capability alone, but unmanaged trust in non-deterministic behavior.

Q: How do security teams know whether NHI governance is actually working?

A: Look for lifecycle completion, not just more inventory. Useful signals include the percentage of identities with owners, the share that are reviewed on schedule, the number of dormant credentials removed, and whether over-privileged access is shrinking in the highest-risk domains.

Q: When should organisations treat an AI system as a non-human identity?

A: Treat an AI system as an NHI when it can authenticate, request tools, or perform actions without direct human supervision. At that point it needs inventory, lifecycle, least privilege, monitoring, and revocation controls just like other machine identities.

Technical breakdown

Why standing AI access becomes a governance problem

AI systems do not become risky only because they are used. They become risky when they are given identities, permissions, and standing access that outlive the task they were meant to perform. In this article, the failure mode is not the presence of AI itself but the speed at which access expands relative to governance. Once an AI system can act inside production systems, every entitlement becomes part of the security boundary, and delayed revocation turns temporary access into persistent exposure.

Practical implication: governance teams need to treat AI access as an entitlement lifecycle problem, not as a policy-only exercise.

How revocation lag changes the blast radius of AI identities

The article highlights a common NHI weakness: organizations may grant access quickly, but they cannot immediately revoke it when the access is no longer needed. That matters because standing access creates a wider blast radius if credentials are misused, over-scoped, or inherited by an agentic workflow. In identity terms, the issue is not just who gets access, but whether the control plane can remove that access at the same speed it was issued.

Practical implication: revocation speed should be measured and tested as a core control, especially for service accounts and AI agent credentials.

Why shutdown and override processes are now part of identity governance

ISACA's survey result shows that many teams do not know how long it would take to halt an AI system or whether they have a documented override process. That is an identity governance problem because control over the system includes the ability to interrupt or constrain what the system can do once it is active. For AI-enabled environments, governance extends beyond provisioning into operational containment, especially where the actor can continue acting unless explicitly stopped.

Practical implication: teams should define and rehearse shutdown authority, override paths, and containment triggers before AI systems are broadly deployed.

Breaches seen in the wild

• LiteLLM PyPI package breach: LiteLLM PyPI supply chain attack, credentials stolen from users.
• Samsung ChatGPT leak 2023: Samsung staff pasted chip source code and meeting notes into ChatGPT weeks after it was allowed, leading Samsung to restrict generative AI tools.

Read and download The State of NHI & AI Agent Breach Report 2026, covering 200+ breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

AI access expansion has become an identity governance test, not an adoption metric. The article shows that organizations are increasing the number of identities with access before they have enough governance maturity to manage those identities. That is especially material for NHI because service accounts and agent credentials behave like production actors, not like passive records. The practitioner conclusion is that access issuance and access governance must be designed as one lifecycle, not two separate programmes.

Standing access is the real control failure behind many AI governance gaps. The cited data points to a structural weakness: organizations can grant access, but they cannot immediately remove it when it is no longer needed. That turns AI and NHI permissions into durable exposure rather than task-bound access. The implication is that access duration, not just access scope, has become a primary governance variable.

AI governance now depends on revocation readiness and shutdown authority. If a team cannot quickly halt an AI system or override its actions, governance is incomplete even when policies exist on paper. This extends the identity model into operational control, because entitlement without interruption capability is only partial governance. Practitioners should treat response readiness as part of identity control design, not as a separate incident playbook.

Ephemeral access is the named concept practitioners should track. The article demonstrates the opposite of ephemeral access: permissions are granted broadly and remain standing after their useful life ends. That creates trust debt across NHIs, agents, and the systems they touch. The practitioner conclusion is to measure whether access truly expires with the task, or merely gets documented after the fact.

AI and NHI governance are converging on the same failure pattern. The same conditions that weaken service-account control also weaken AI access control: broad entitlements, slow revocation, and uncertain shutdown paths. That means identity teams should stop treating AI governance as a separate conversation from NHI lifecycle management. The practitioner conclusion is to govern AI as a non-human identity class with explicit lifecycle and containment controls.

From our research library:

• Systems with least-privileged AI access had a 17% incident rate vs 76% for over-privileged systems. Organisations failing to scope AI access properly are 4.5x more likely to experience a security incident, according to the 2026 Infrastructure Identity Survey.
• 70% of organisations grant AI systems more access than they would give a human employee performing the exact same job, according to the 2026 Infrastructure Identity Survey.
• Read next: NHI Lifecycle Management Guide

What this signals

Ephemeral access is becoming the dividing line between AI governance that exists on paper and governance that actually contains risk. When access can be granted broadly and removed only after delay, entitlement lifetime becomes the security problem. That makes lifecycle ownership, not policy count, the metric that matters for AI and NHI programmes.

AI access control now needs the same discipline as non-human identity lifecycle management. Organizations that cannot inventory, scope, and revoke machine access in step with deployment will keep creating standing privilege in the background. The practical signal is simple: if revocation cannot happen immediately, access was never truly temporary.

For practitioners

• Measure access revocation speed Test how long it takes to remove standing access from service accounts, tokens, and AI agent credentials after the entitlement is no longer needed.
• Define AI shutdown authority Document who can stop or override an AI system during an incident, and verify that the authority path works across operations and security teams.
• Scope non-human identity governance coverage Inventory which service accounts, bot accounts, and agent credentials are actually governed, then close the gap where access exists without lifecycle ownership.
• Separate policy from containment Treat a written AI policy as incomplete until it is paired with documented override procedures, access bounds, and tested revocation steps.

Key takeaways

• AI access expansion is exposing a governance gap between issuance and revocation, and that gap maps directly to higher breach exposure.
• The article's evidence points to weak NHI coverage, delayed removal of standing access, and unclear shutdown readiness as the core failure conditions.
• Practitioners should measure revocation speed, document shutdown authority, and govern AI systems through the same lifecycle lens used for other non-human identities.

Key terms

• Non-Human Identity (NHI): A digital identity assigned to a non-human entity such as a software application, service account, API key, bot, machine, or AI agent that enables it to authenticate and interact with systems without direct human involvement. NHIs now outnumber human identities in most enterprises by 25 to 50 times.
• Standing Access: Standing access is persistent privilege that remains available without fresh approval or contextual checks. In NHI environments, standing access usually appears as long-lived tokens, reusable service accounts, or broad roles attached to automation. It is convenient operationally, but it expands risk when conditions change or secrets leak.
• Revocation Latency: Revocation latency is the time between a decision to remove access and the point at which that access is actually gone. It is a practical measure of how long stale privilege remains usable after a role change, offboarding, or contract end. Shorter latency means smaller exposure and cleaner audit evidence.
• Shutdown Authority: Shutdown authority is the documented ability to stop, override, or contain a running system when its behaviour becomes unsafe. For AI-enabled environments, it is part of operational governance because policies alone do not constrain an active system unless a real interruption path exists.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an IAM programme, it is worth exploring.