AI data security controls are lagging behind enterprise AI use

At a glance

What this is: NHI Mgmt Group's analysis, drawing on Cyera's research: Enterprise AI use outpacing data security controls, with visibility, real-time monitoring, and automatic blocking all lagging behind adoption.

Why it matters: For IAM and security teams, the issue is not AI adoption itself but governable access, because weak visibility and over-permissioning create identity and data exposure across AI workflows.

By the numbers:

• 83% of enterprises already use AI in daily work, according to Cyera research.
• Only 13% report strong visibility into how AI touches enterprise data, according to Cyera research.
• 76% say autonomous AI agents are the hardest to secure, according to Cyera research.
• Only 11% can automatically block risky AI activity, according to Cyera research.

Context

Enterprise AI is creating an identity and data governance gap: usage is now widespread, but security teams still lack reliable visibility into how AI interacts with sensitive information. The result is a control model that assumes AI can be managed like a normal application or user, even when its prompt-driven behaviour and broad access patterns break that assumption.

Cyera’s survey frames this as a readiness problem rather than an adoption problem. When organisations cannot see prompt-layer activity, cannot monitor AI in real time, and cannot automatically block risky actions, they are governing AI after the fact instead of at the point of access.

That gap matters most when AI becomes a decisioning layer between people and data. In those cases, access scope, output handling, and governance ownership all become identity questions as much as data-security questions.

Key questions

Q: What breaks when AI agents are treated like standard human users?

A: You lose visibility into effective permissions, expected behaviour, and real blast radius. Human-centric controls can misclassify normal agent activity as compromise, or miss policy violations that happen entirely within legitimate access. The failure is not only technical, it is governance design that assumes a person is always behind the action.

Q: Why do AI agents create a bigger governance problem than ordinary endpoint tools?

A: Because an AI agent can execute many file reads, API calls, and transfers in one session without a human approving each step. That collapses the assumption that access is reviewed before action. Governance must therefore focus on delegated machine activity, not only on user approval flows.

Q: How can organisations tell whether their AI security model is actually working?

A: They should test whether the control stack can explain who acted, what data was touched, and what purpose the action served. If those three signals cannot be correlated in one incident view, the model is likely monitoring access without governing behaviour. That is a visibility gap, not a complete AI security posture.

Q: Should organisations use the same access model for humans and AI agents?

A: No. Human access models are built around stable roles and review cycles, while AI agents often need contextual, task-specific permissions that change quickly. Treating them the same usually leads to over-permissioning or constant exceptions. Organisations should separate identity proof from authorization design and apply resource-level controls for agents.

Technical breakdown

Prompt-layer visibility is the missing control plane

Prompt-layer visibility means capturing what an AI system was asked, what data it touched, and what it returned in time to act on it. The article’s core finding is that organisations often have neither real-time telemetry nor enough auditability to understand AI behaviour at the point it matters. That makes prompt and output logging more than a forensic feature. It becomes the only way to see whether AI is operating within approved data boundaries or quietly expanding exposure through ordinary business use.

Practical implication: instrument prompts, outputs, and policy decisions before AI use becomes too distributed to observe.

Autonomous AI agents change the governance problem

An autonomous AI agent is not just another workload. It can select actions and timing dynamically, which means security teams are no longer governing a fixed access path but an execution path that can change mid-task. That is why respondents singled out autonomous agents as the hardest to secure. Once the system can decide what to do next, static entitlement reviews and quarterly access checks no longer describe the real risk window.

Practical implication: treat agent-driven execution as a live governance problem, not a periodic review problem.

AI access should be governed as an identity and privilege issue

The article shows that many programmes still treat AI as if it sits outside identity governance, even though it consumes data through credentials, permissions, and policy enforcement points. If AI can over-access sensitive information, then the control failure is not only data classification. It is the mismatch between what the system is allowed to reach and what its task actually requires. That is the same basic governance problem seen in overprivileged NHIs, just expressed through an AI interface.

Practical implication: bind AI access to least privilege, lifecycle ownership, and data classification instead of broad default access.

NHI Mgmt Group analysis

AI governance is becoming an identity problem before it becomes a data problem. Cyera’s survey shows that enterprises are already using AI at scale while only a small minority can see how it touches data in real time. That means the first failure is not exfiltration, but the absence of a governable identity boundary around AI activity. Practitioners should treat AI access as a first-class entitlement model, not a side effect of application rollout.

Prompt-layer control is the new policy boundary. The article repeatedly points to prompt and output handling as the place where risk appears and where controls are missing. Traditional data security can classify sensitive content, but it cannot explain why an AI system should have seen it in the first place. The field now needs policy that governs the interaction point, not just the data store.

Autonomous AI agents expose a runtime governance gap that static reviews cannot close. Access review processes were designed for access that persists long enough to be reviewed. That assumption breaks when an AI agent can request, use, and release privileges as part of a single task. The implication is that governance must move closer to issuance and runtime enforcement, because after-the-fact certification no longer matches the behaviour being governed.

Over-access in AI is the same structural failure as NHI privilege drift. When 66% of organisations report AI over-accessing sensitive information, the problem is not novelty but reuse of an old governance pattern in a new context. AI systems are being granted data reach that exceeds task need, just as service accounts often outgrow their original scope. The practitioner conclusion is straightforward: identity scope, not model sophistication, is what determines blast radius.

Named concept: prompt-layer readiness gap. This article shows a specific gap between AI adoption and the controls needed to supervise the interaction layer where prompts, outputs, and permissions intersect. The more organisations rely on AI for everyday work, the more this gap shifts from a monitoring issue to a programme design issue. Teams should map governance to observable evidence, not just stated policy.

From our research library:

• 53% of security leaders expect AI to run major portions of their infrastructure autonomously within the next three years, according to the 2026 Infrastructure Identity Survey.
• 19% of organisations give AI systems dramatically more access than human employees, nearly one in five granting unrestricted privilege, according to the 2026 Infrastructure Identity Survey.
• Read next: AI Agent Identity Security Buyer's Guide

What this signals

Prompt-layer readiness gap: AI programmes now fail or succeed at the interface where prompts, outputs, and entitlements meet. Organisations that cannot observe that layer in real time will keep discovering AI risk only after access has already been exercised.

Autonomous agents make governance time-sensitive in a way standard recertification cycles were never designed to handle. Access can be acquired, used, and discarded within one task, which means the control point has to move closer to issuance and runtime enforcement.

Use the early AI programme to define whether AI is a governed identity class, a data consumer, or an uncontrolled extension of user access. That decision shapes how you set ownership, approvals, and revocation paths across the rest of the programme.

For practitioners

• Instrument AI activity at the prompt layer Capture prompts, outputs, access decisions, and anomaly signals early enough to support containment rather than only post-incident review.
• Assign AI a first-class identity model Treat AI systems as governed identities with named owners, scoped permissions, and explicit lifecycle control rather than as informal extensions of user access.
• Constrain autonomous agent scopes Limit the data and tool reach of autonomous agents to the smallest task-specific boundary, and require approval gates for higher-risk actions.
• Make over-access visible Review where AI is granted broader data access than equivalent human roles and reduce standing privilege wherever the task does not justify it.
• Tie governance to evidence Define measurable controls for visibility, block rates, and incident containment so policy can be tested against actual AI behaviour.

Key takeaways

• AI use is already widespread, but governance still trails the way AI touches data in practice.
• The most material gap is visibility at the prompt and access layer, where over-permissioning and weak monitoring combine.
• Teams need identity-style controls for AI, because runtime access scope determines the blast radius more than model sophistication does.

Key terms

• Interaction-Layer Visibility: The ability to see prompts, responses, account context, and resulting actions as one governed event. This is the control layer where AI usage becomes auditable, because traffic monitoring alone cannot explain whether a prompt came from an approved account or how the session behaved.
• Autonomous AI Agent: An autonomous AI agent is software that can perceive inputs, decide what to do, and act with limited or no human prompting. In identity security, it is treated as a non-human identity when it can authenticate, call tools, access data, or trigger workflows under its own runtime decisions.
• First-Class Identity Object: A first-class identity object is a governed record that exists independently of the credential it uses. For AI agents, this means the enterprise can tie access, ownership, and review obligations to the actor itself rather than to a service account, API key, or application label.
• Over-Permissioning: Over-permissioning occurs when an identity receives more access than it needs to complete its assigned work. It is often introduced as a convenience to avoid creating new roles or handling exceptions, but it expands attack surface and creates unnecessary governance risk.

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.