AI in cybersecurity is useful for context, not control

At a glance

What this is: This analysis argues that AI improves cybersecurity context and triage, but control still depends on deterministic enforcement, not probabilistic inference.

Why it matters: For IAM, NHI, and security teams, the key lesson is that AI can support decision-making, but privilege, access, and containment still need enforceable controls.

By the numbers:

• 99% of U.S. adults engage with AI-enabled products weekly, according to Zero Networks.
• 64% of people don’t know they’re using AI features, according to Zero Networks.
• over 70% of security leaders have adopted or are evaluating AI for their security operations, according to Zero Networks.
• 83% say they're overwhelmed by alert volume, false positives, and a lack of context, according to Zero Networks.

Context

AI security tools are often framed as control mechanisms, but most current use cases are better understood as decision support. They can summarise telemetry, rank alerts, and surface patterns, yet those functions do not by themselves enforce least privilege, stop movement, or contain an active compromise. That distinction matters most in environments where identity, access, and response must be deterministic.

The article sits at the intersection of cybersecurity operations and identity governance because the most consequential failures appear when probabilistic AI is mistaken for an enforcement layer. In NHI and IAM programmes, that mistake can turn AI into an observation layer that is useful for triage but irrelevant to actual privilege control.

The starting position described here is now typical: AI is embedded in security workflows, but teams still struggle to separate context generation from control execution.

Key questions

Q: What breaks when AI is used as a security control instead of a decision aid?

A: The main failure is that AI can identify risk without preventing it. If the environment depends on AI to act as the control, then suspicious access, lateral movement, and privilege abuse may continue until a human intervenes. Control still has to come from deterministic enforcement such as policy, segmentation, and revocation.

Q: Why do AI-driven security tools still need deterministic enforcement?

A: Because AI produces probabilistic judgments, not guaranteed outcomes. Security teams need controls that can deny access, terminate sessions, or isolate systems based on defined conditions. Without that enforcement layer, AI may improve triage while leaving the underlying exposure unchanged.

Q: How can security teams tell whether AI is only improving visibility, not containment?

A: If AI improves dashboards, alert ranking, or analyst workflow but does not change access decisions or block attack paths, it is only improving visibility. Teams should look for automated actions such as revocation, isolation, and policy enforcement to prove that AI is affecting containment, not just reporting.

Q: How should organisations govern digital identity when AI is part of the service model?

A: They should treat identity assurance, access control, and AI oversight as one governance chain. If AI is making or influencing decisions in regulated services, the organisation needs auditable identity evidence for the initiating actor, the permissions used, and the policy that allowed the action. Without that linkage, AI governance is hard to defend.

Technical breakdown

Probabilistic AI vs deterministic control in security operations

AI systems infer likely meaning from patterns, confidence scores, and prior examples. Deterministic controls, by contrast, apply explicit rules to known conditions, such as denying a connection, limiting a privilege, or segmenting a network path. In security operations, that difference determines whether a tool merely recommends action or actually changes the system state. AI can improve analyst judgment, but it cannot prove enforcement on its own. In identity-heavy environments, that gap is critical because access decisions need auditable logic, not a best guess.

Practical implication: Use AI for analysis and routing, but anchor enforcement in deterministic access and network controls.

Why AI helps SOC teams but does not replace SIEM, EDR, or XDR

Security telemetry platforms already collect high-volume event data, but the practical challenge is turning noise into an actionable sequence. AI is strongest when it compresses logs, correlates signals, and helps analysts prioritise what matters first. That is different from making the SIEM, EDR, or XDR stack itself authoritative. The enforcement layer still depends on defined detections, response playbooks, and control actions that can be executed without ambiguity. Where AI is treated as the response engine, the organisation risks faster interpretation but slower containment.

Practical implication: Keep AI above the detection stack as a triage layer, not below it as the control plane.

Why AI cannot govern standing access or lateral movement by itself

The article’s central limitation is that AI observes suspicious behaviour, but it does not inherently revoke credentials, narrow privilege scope, or stop a session from expanding. That matters for service accounts, privileged users, and network paths because lateral movement is prevented by access design, not by pattern recognition alone. In other words, the security question is not whether AI can notice movement, but whether the architecture makes movement impossible or short-lived when it should be. Identity governance is still the control point.

Practical implication: Pair AI-driven detection with least privilege, segmentation, and revocation paths that act automatically.

Threat narrative

Attacker objective: The attacker aims to move faster from access to privilege expansion and reach high-value systems before defenders can contain the session.

1. Entry begins with AI-assisted phishing, reconnaissance, or targeting that increases the quality and speed of initial access attempts.
2. Escalation follows when attackers use AI to identify exposed services, weak configurations, or privilege escalation opportunities inside the environment.
3. Impact occurs when defenders rely on AI to detect suspicious behaviour but lack deterministic controls to stop lateral movement or revoke access.

Breaches seen in the wild

• Sisense breach 2024: A credential in Sisense's GitLab reportedly opened S3 buckets of customer tokens, passwords and certificates; CISA urged a full reset.
• CISA Private-CISA GitHub leak 2026: A CISA contractor's public GitHub repo exposed AWS GovCloud admin keys, Artifactory credentials and plaintext passwords for six months.

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 is a context layer, not a control layer. The article correctly separates summarisation and prioritisation from enforcement, which is where many security programmes still blur the line. AI can help analysts decide faster, but it does not establish whether access should exist, continue, or be revoked. The practitioner conclusion is that AI belongs in the decision-support stack, while enforcement remains the job of deterministic controls.

The real security risk is control substitution. Organisations create governance debt when they let probabilistic inference stand in for revocation, segmentation, or privilege restriction. That risk is especially acute in NHI and identity-heavy environments, where access decisions must be precise and auditable. The named concept here is the context-control gap: the widening distance between what AI can explain and what the environment can actually enforce. Practitioners should treat that gap as a design flaw, not a maturity issue.

Explainability is a governance requirement, not a product feature. If AI decisions cannot be explained to a board, regulator, or incident responder, they cannot carry operational authority. This is particularly relevant where AI touches access decisions, security prioritisation, or incident routing. The practical conclusion is that AI outputs need traceability, but the underlying control decision still has to be independently defensible.

AI-assisted attackers increase the value of containment design. The article’s attacker section shows that speed is now part of the threat model, from phishing to lateral movement. That means detection latency matters less if the architecture already constrains how far an attacker can go. The practitioner conclusion is that resilience comes from reducing the blast radius, not from asking AI to recognise compromise sooner.

Identity governance remains the hard boundary. When the article notes that AI cannot revoke access or stop movement, it is pointing to the place where IAM, PAM, and NHI governance must stay in charge. AI can surface the problem, but only identity controls can bound it. The practitioner conclusion is that AI should inform identity decisions, not inherit them.

What this signals

Context-control gap: the strongest analytical warning in this article is that security teams can mistake visibility for containment. AI may improve prioritisation, but the environment still needs policy enforcement that can deny, isolate, or revoke without waiting for human interpretation.

When AI is introduced into identity-heavy security programmes, the practical test is whether a recommendation becomes an action. If the answer is no, the programme has added another layer of intelligence without changing blast radius, which is a poor trade when adversaries are also using AI to move faster.

For practitioners

• Separate AI analysis from enforcement Use AI to summarise telemetry, prioritise alerts, and cluster incidents, but keep revocation, blocking, and segmentation in deterministic control paths.
• Bind AI outputs to actionable control points Connect AI-driven findings to controls that can actually change access, such as conditional access, session termination, or network isolation.
• Review privileged and non-human access paths Check whether service accounts, workloads, and admin sessions still rely on human review after the fact instead of automatic enforcement at issuance time.
• Require explainability for AI-assisted decisions Document how the system reached each recommendation so analysts, managers, and auditors can trace why a risk score or alert ranking changed.

Key takeaways

• AI adds value when it improves context, summarisation, and alert prioritisation, but those functions do not equal control.
• The operational risk is using probabilistic outputs where deterministic enforcement is required for access, containment, and privilege management.
• Security teams should treat AI as an input to governance decisions and keep revocation, segmentation, and blocking in the control plane.

Key terms

• Deterministic Controls: Deterministic controls are rules that produce predictable, repeatable outcomes for an AI agent’s actions. They remove ambiguity by enforcing predefined conditions, allowed paths, and stop points. In security testing, they help ensure the agent behaves consistently, stays within scope, and produces results that can be audited and trusted.
• Probabilistic Security: A security approach that relies on likelihood, scoring, or inference rather than fixed rules. It is useful for prioritisation and detection, but it cannot guarantee enforcement because the system is estimating risk rather than making a hard authorisation decision.
• Context-Control Gap: The context-control gap is the distance between understanding a security problem and actually enforcing a response. AI can narrow that gap by improving insight, but the gap remains whenever analysis does not translate into blocking, revocation, or containment.
• Deterministic Automation: Deterministic automation is policy execution that produces predictable results without improvisation or model-driven guesswork. In resilience programmes, it matters because containment must happen consistently under pressure, with clear auditability and minimal human delay.

Deepen your knowledge

NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security, and secrets management. It helps practitioners align identity controls with broader security operations and AI-driven workflows.