Identity security is shifting from visibility to enforcement at RSAC

At a glance

What this is: This blog argues that identity security is shifting from seeing access to enforcing it, especially as AI agents and non-human identities become the dominant privileged actors.

Why it matters: It matters because IAM, PAM, and governance teams now have to enforce policy at runtime across human, NHI, and agentic access paths instead of relying on visibility, vaulting, or human-centric assumptions.

By the numbers:

• Through 2028, at least 80% of unauthorized AI agent transactions will be caused by internal violations of enterprise policies concerning information oversharing, unacceptable use or misguided AI behavior rather than from malicious attacks.
• Only 5.7% of organisations have full visibility into their service accounts.
• 97% of NHIs carry excessive privileges, increasing unauthorised access and broadening the attack surface.

👉 Read Silverfort's RSAC identity security Q&A on enforcement and AI agent governance

Context

The core problem is not identity visibility alone, but the governance gap between knowing what an identity can do and enforcing what it should be allowed to do. In the RSAC discussion, the emphasis is on identity-first control because traditional security models were built around human-paced administration, not AI agents or NHIs that can act and disappear quickly.

That shift matters for IAM and PAM teams because the control point is moving from inventory and review to runtime enforcement. If access can be requested, selected, and exercised in seconds, then post-hoc visibility cannot carry the governance burden by itself; policy has to travel with the identity at the moment of action.

Key questions

Q: How should security teams enforce privileged access in hybrid environments?

A: Security teams should enforce privileged access as a runtime decision, not just as a stored credential problem. That means binding policy to authentication and action time, then checking whether the same control path applies across cloud, SaaS, automation, and AI-driven identities. If a privilege can be used without a policy decision at the moment of use, the programme still depends on visibility rather than enforcement.

Q: Why do vaults alone not solve privileged access risk?

A: Vaults protect where credentials live, but they do not control how those credentials are used once an identity is active. Privilege risk often emerges at execution time, inside hybrid and agentic workflows that never depend on the vault after checkout. Teams need controls that govern use, not only storage, if they want actual privilege reduction.

Q: What do organisations get wrong about AI agent governance?

A: They often treat AI agents like scripted automation or like human admins with faster output. That misses the governance problem: an agent can make decisions and act at runtime, so approval loops, static reviews, and human-paced control cycles do not reliably contain the risk. Governance has to define acceptable behaviour and enforce it where the action happens.

Q: How do you know if identity visibility is actually improving security?

A: Visibility is only improving security if it leads to fewer uncontrolled access paths and more decisions enforced at the point of use. If discovery produces better reports but no change in runtime policy, the programme has improved knowledge, not control. The best indicator is whether the same identity that can be seen can also be constrained before it acts.

Technical breakdown

Why visibility does not equal enforcement in identity security

Visibility tells you which identities exist, what they can access, and where they appear in logs. Enforcement is different: it decides whether an action is allowed at the moment of authentication or request execution. In environments with AI agents and NHIs, that distinction matters because the identity may exist only briefly, act quickly, and leave little reviewable history behind. A mature identity programme therefore has to bind policy to runtime access decisions, not just to periodic discovery or reporting.

Practical implication: Practitioners should treat discovery as input to enforcement design, not as a substitute for it.

Why vault-based PAM is insufficient for modern privileged access

Vaults protect stored credentials, but they do not by themselves control how credentials are used once an identity is active. That is the weak point in a privilege model built mainly around secret storage. In hybrid estates, privilege is also distributed across cloud, SaaS, automation, and agentic workflows, so a control that only hides credentials leaves too much of the access path ungoverned. Inline controls at authentication and access time close that gap more directly than storage-centric models.

Practical implication: Use vaulting as one control layer, but anchor privileged access decisions in runtime policy enforcement.

How AI agent security changes the identity problem

AI agents change identity governance because they can behave like independent executors rather than static service accounts. The control problem is no longer only about proving who or what the identity is, but about constraining what it may decide to do after authentication. If the agent can select actions and execute tasks at speed, then traditional human approval loops and scheduled recertification do not map cleanly to the runtime risk. That is why policy definition, acceptable-use boundaries, and enforcement timing become central to agent governance.

Practical implication: Security teams should design access controls for runtime decision points, not just for onboarding and review.

Breaches seen in the wild

• Moltbook AI agent keys breach — Moltbook breach exposed 1.5M AI agent keys.
• Cisco DevHub NHI breach — IntelBroker exploited exposed Cisco credentials, API tokens and keys in DevHub.

Read our 52 NHI Breaches Analysis report for a comprehensive view of breaches impacting Non-Human Identities including AI Agents.

NHI Mgmt Group analysis

Identity control is replacing identity visibility as the strategic centre of gravity. The article’s core argument is that organisations already know they need more visibility, but visibility alone does not reduce risk. Once identities can be discovered, the real question becomes whether policy can be enforced before action. That is a structural shift in IAM and PAM programmes, not a reporting improvement, and practitioners should read it as a mandate to move from observation to control.

Vault-centric PAM is no longer sufficient for a hybrid privileged landscape. Vaults solve credential storage, not the runtime misuse of privileged access across humans, NHIs, and agentic workflows. The article correctly points to privileged activity spreading beyond Tier-0 and across hybrid environments, which makes selective, storage-only coverage too narrow. The implication is that privileged access governance must follow the action path, not just the secret.

AI agents are forcing identity governance to absorb a new actor type. Calling them virtual employees is useful only if the control model changes with it. AI agents are not static automations, and governance that assumes a human operator or a long-lived service account will misread their behaviour. Practitioners should treat this as a category shift, not a branding change, because the decision logic now lives inside the actor.

Runtime enforcement is the named control gap that matters most here. The article points to a runtime governance gap, which is the difference between knowing access exists and being able to stop misuse at the point of action. That gap affects OWASP-NHI, ZT-NIST-207, and NIST-CSF interpretations of least privilege and continuous verification. The practitioner takeaway is to re-evaluate whether current controls actually intervene at execution time.

From our research:

• Only 5.7% of organisations have full visibility into their service accounts, according to Ultimate Guide to NHIs.
• Our research also shows that 97% of NHIs carry excessive privileges, which broadens the attack surface and makes runtime enforcement harder to defer.
• For a wider view of where these failures show up in practice, the 52 NHI Breaches Analysis shows how privilege and visibility gaps translate into real compromise patterns.

What this signals

The next planning cycle for IAM and PAM teams should assume that visibility projects will not be judged on inventory quality alone. The market is moving toward runtime control, and programmes that cannot enforce policy at access time will be seen as incomplete even if their discovery coverage looks strong.

Runtime governance gap: the real programme gap is no longer whether identities are known, but whether their actions are controlled at the moment of use. Teams that can align privileged access, AI agent policy, and zero-trust enforcement around that gap will be better positioned to manage mixed identity estates.

If your organisation is still relying on vaulting as the centre of privileged control, the immediate signal is to evaluate whether that model covers agentic access paths at all. NIST Cybersecurity Framework 2.0 is a useful reference point for tying identity controls back to govern, protect, detect, respond, and recover functions.

For practitioners

• Move from visibility goals to enforcement goals Reframe identity programme success around whether policy is enforced at authentication and access time, not whether identities are merely inventoried and reported on.
• Map privileged access beyond vaults Review where privileged decisions occur in cloud, SaaS, automation, and agentic workflows, then identify the places where vault-based controls never see the action path.
• Define acceptable use for AI agents Create explicit policy boundaries for agent behaviour, including what actions are allowed, what data may be accessed, and which execution paths require tighter control.
• Test runtime control coverage across all identities Validate whether human admins, service accounts, and AI agents are all subject to the same enforcement logic, or whether one class still escapes real-time policy.

Key takeaways

• Identity security is moving from observation to enforcement, and that changes the control model rather than just the tooling stack.
• Vaults improve credential protection, but they do not by themselves govern runtime privilege use across hybrid identities and AI agents.
• Teams that define policy at the moment of access will be better positioned than those that depend on post-hoc visibility and review.

Key terms

• Runtime enforcement: Runtime enforcement is the practice of deciding whether an identity may act at the moment access is requested or used. In identity security, it matters because visibility alone cannot stop misuse. For AI agents and NHIs, enforcement has to operate fast enough to constrain actions before the actor completes them.
• Vaultless PAM: Vaultless PAM is a privileged access model that reduces reliance on credential vaulting as the main control point. Instead of protecting secrets and hoping that checkout use is safe, it enforces access decisions inline at authentication and authorization time. That makes it more aligned to hybrid and fast-moving identity behaviour.
• AI agent identity: AI agent identity is the identity assigned to a software entity that can make independent runtime decisions, select tools, and execute actions without human approval gates. Unlike scripted automation, it introduces governance questions around acceptable behaviour, action scope, and timing because the actor can change what it does while a session is still active.
• Identity-first security: Identity-first security is an approach that treats identity as the primary control plane for managing risk. Instead of relying mainly on network or endpoint boundaries, it uses identity context to decide what can happen, when it can happen, and under what conditions. That model is especially relevant where privileges move across human, non-human, and agentic actors.

Deepen your knowledge

Identity security enforcement, privileged access governance, and AI agent policy are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are redesigning controls around runtime decision points, it is a practical place to start.

This post draws on content published by Silverfort: RSAC Conference 2026 identity security takeaways and Q&A with Abbas Kudrati. Read the original.