AI governance and continuous compliance are reshaping GRC priorities

At a glance

What this is: Clarity Security says ISACA conference sessions showed GRC leaders shifting from checkbox compliance toward continuous risk reduction, with AI governance and evidence generation now central concerns.

Why it matters: For IAM, NHI, and GRC teams, the message is that point-in-time reviews are no longer enough when systems can expand access, make decisions, and leave incomplete audit evidence.

By the numbers:

• The ISACA GRC conference brought together over 1500 compliance, audit, technology, and security leaders in San Diego.

👉 Read Clarity Security's analysis of AI governance and continuous compliance at ISACA

Context

AI governance is moving from a future concern to an operating issue for security, compliance, and audit teams. The article frames this through ISACA conference discussions that linked AI, agentic systems, MCP risk, and continuous compliance to a broader breakdown in point-in-time control models. For identity and governance teams, the implication is straightforward: if systems can alter access and decisioning faster than review cycles, the control model needs to shift with them.

The article’s GRC framing is especially relevant to identity programmes because access reviews, segregation of duties checks, and evidence generation all depend on visibility into who or what has access. That includes human users, non-human identities, and increasingly AI-driven systems that may act without a stable, human-readable workflow. The starting position described here is typical of most enterprises, not exceptional.

The article’s GRC framing is especially relevant to identity programmes because access reviews, segregation of duties checks, and evidence generation all depend on visibility into who or what has access. That includes human users, non-human identities, and increasingly AI-driven systems that may act without a stable, human-readable workflow. The starting position described here is typical of most enterprises, not exceptional.

Key questions

Q: What breaks when compliance is still run as a point-in-time exercise?

A: Point-in-time compliance breaks when the environment changes faster than the review cycle. The result is defensible paperwork that no longer matches live access, system behavior, or control status. That creates false assurance, delayed remediation, and audit evidence that can be technically complete but operationally stale.

Q: Why does agentic AI create both security gains and governance risk in incident response?

A: Agentic AI creates value because it can process large datasets quickly, spot anomalies, and act before damage spreads. The same autonomy creates governance risk because decisions can be made without direct human review, which raises concerns about explainability, bias, and unintended remediation. Teams need clear policies, tested controls, and monitoring so speed does not outpace accountability.

Q: How do organisations know if continuous compliance is actually working?

A: Continuous compliance is working when evidence is current, exceptions are visible, and remediation is tracked in the same workflow as the control. If teams still need large manual evidence-gathering exercises before audits, the programme is still periodic at heart. The strongest signal is that access and control status can be verified at any time.

Q: What should GRC and IAM teams do when audit evidence does not match effective access?

A: Treat the mismatch as an operating-control failure, not a documentation issue. Investigate the source of truth for identity, entitlement, and system access, then fix the lifecycle gap that lets reported access diverge from real access. The goal is to restore evidence that reflects what users, services, and agents can actually do.

Technical breakdown

Why point-in-time compliance fails in continuous environments

Point-in-time compliance assumes the state being reviewed is still true when the audit evidence is produced. That breaks down when access, entitlements, and system behavior change continuously across cloud, SaaS, and AI-enabled workflows. In practice, a clean review can coexist with active risk minutes later. The problem is not auditing itself, but the mismatch between static verification and dynamic control surfaces. For identity teams, the challenge extends beyond human access to service accounts, delegated permissions, and agentic systems that do not stay still long enough for legacy review cadences to capture them.

Practical implication: Shift from snapshot-based review thinking to continuous control verification for identities, entitlements, and AI-driven access paths.

MCP and agentic AI create a new governance surface

Model Context Protocol connects AI agents to tools and data sources, which means governance now has to account for delegated action, tool invocation, and runtime access scope. When an agent can request information, call services, or trigger workflows, the control question is no longer just whether the model is allowed to exist. It becomes what it can reach, when, and under whose authority. That turns AI governance into an identity problem as much as a model-risk problem, because every tool link effectively creates another access boundary that must be classified, monitored, and bounded.

Practical implication: Treat MCP-connected agents as governed identities with scoped access, not as generic software components.

Evidence generation now depends on complete identity visibility

Audit evidence fails when the organisation cannot reconcile reported access with effective access. That gap widens when environments include nested identities, federated users, non-human identities, and AI-mediated access flows. Manual evidence collection becomes brittle because the underlying data is fragmented across systems, teams, and control owners. The deeper issue is that GRC programmes often assume identity inventories are stable enough to certify after the fact. In a continuous environment, evidence has to be assembled from live control signals, not reconstructed by hand from inconsistent snapshots.

Practical implication: Build evidence workflows around live identity and access telemetry so audits can verify current state instead of reconstructing it.

Threat narrative

Attacker objective: The objective is to exploit governance blind spots to gain or extend access while leaving compliance evidence that looks valid but no longer matches operational reality.

1. Entry begins when AI systems connect to tools and data through MCP-style integrations, expanding the attack and governance surface beyond the model itself.
2. Credential and access abuse follows when static or poorly scoped permissions let agents, users, or delegated workflows reach more data and systems than intended.
3. Impact appears as audit gaps, segregation-of-duties failures, and control evidence that no longer reflects the real access state of the environment.

NHI Mgmt Group analysis

Checkbox compliance is becoming a false comfort layer. The article’s core insight is that passing audits does not mean reducing risk when access and system behavior change continuously. GRC programmes that optimise for evidence completion rather than control effectiveness will keep producing neat artifacts and unresolved exposure. The practitioner conclusion is that compliance has to become a byproduct of continuous risk control, not the primary objective.

AI governance is now an identity governance problem. Once AI systems can make decisions, call tools, and influence workflows, the relevant control unit is no longer only the model. It is the delegated identity, the tool boundary, and the permission scope attached to that runtime behavior. That is why governance frameworks built for static applications struggle here. The practitioner conclusion is that AI oversight must include identity lifecycle, access scope, and runtime authorization.

Evidence generation is the new bottleneck in GRC maturity. The article shows that organizations do not only lack control, they often lack defensible proof that controls are working. That is a structural weakness because regulators increasingly expect current-state evidence, not historical summaries. This is the same pattern identity teams face when inventories lag reality. The practitioner conclusion is to treat evidence quality as an operational control, not an afterthought.

Continuous compliance will force a redesign of review cadences and control ownership. The traditional split between audit, security, and IAM no longer works when the environment changes faster than any one team can certify it. The article points to a future where governance is continuous, cross-functional, and instrumented through live telemetry. The practitioner conclusion is to align access review, AI governance, and audit evidence around one operating model.

From our research library:

• 67% of organisations still rely heavily on static credentials despite the risks they pose to agentic AI deployments, according to the 2026 Infrastructure Identity Survey.
• Only 13% of organisations feel extremely prepared for the reality of agentic AI despite the majority racing toward autonomous adoption, according to the 2026 Infrastructure Identity Survey.
• Read next: Access Reviews and Certification Guide

What this signals

Continuous compliance is becoming an identity visibility problem. When access reviews, delegated permissions, and non-human identities all change faster than audit cycles, the programme has to move from retrospective attestation to live verification. That is especially true where AI agents and MCP-connected tools introduce access paths that traditional review templates do not model.

Evidence gaps will increasingly expose NHI and agentic AI governance debt. If an organisation cannot reconcile reported access with effective access, it cannot prove control effectiveness with confidence. That makes identity inventory quality, entitlement telemetry, and lifecycle governance the practical foundation for GRC modernisation.

67% of organisations still rely heavily on static credentials despite the risks they pose to agentic AI deployments, according to the 2026 Infrastructure Identity Survey. That figure matters here because continuous compliance collapses when the underlying access model is still static while the operating model is becoming autonomous.

For practitioners

• Rebuild compliance around continuous control validation Replace point-in-time evidence collection with ongoing checks that confirm access, SoD, and configuration state while systems are live.
• Classify AI systems as governed access paths Map agentic AI, MCP-connected tools, and delegated workflows to the identities, permissions, and approvals they actually consume.
• Reconcile reported access with effective access Compare what IAM and GRC systems say should exist with what databases, SaaS platforms, and non-human identities can truly do.
• Instrument audit evidence from live telemetry Pull access, entitlement, and control signals directly from production systems so evidence reflects the current operating state rather than a manual reconstruction.

Key takeaways

• AI governance is no longer separate from identity governance when systems can make decisions and invoke tools under delegated authority.
• The main failure mode is not the absence of audits but the mismatch between live access reality and the evidence used to prove control.
• GRC teams should redesign evidence, review cadence, and access visibility around continuous verification instead of retrospective certification.

Key terms

• Continuous Compliance: Continuous compliance is the practice of keeping controls and evidence current as the environment changes, rather than proving compliance after a review cycle. For identity and NHI programmes, it means access, logging, and revocation must operate together in real time.
• Agentic AI: Autonomous AI systems capable of planning, deciding, and taking actions, including calling APIs, writing code, and orchestrating other agents, with minimal human oversight. Agentic AI introduces new NHI risks as agents must authenticate to external services.
• Effective Access: The actual permissions an identity can exercise after inheritance, nested groups, delegation, and object-level controls are evaluated. In Active Directory, effective access is more useful than direct membership because it reveals the true operational reach of a service account.
• Model Context Protocol: Model Context Protocol is an open protocol that lets AI agents connect to tools and data sources. It expands what an agent can reach, so governance has to cover not only the model and its prompts, but also every system that can receive or return agent-driven data.

What's in the full article

Clarity Security's full article covers the conference themes and operational implications this post intentionally leaves at a higher level:

• Specific session themes on AI governance, continuous compliance, and evidence generation from the ISACA conference floor
• The practical discussion around MCP risk and why it changes how teams think about agentic AI access
• Examples of how conference speakers framed audit methodology when systems can make their own decisions
• The article's own view on why point-in-time assessments struggle to produce defensible evidence

👉 Clarity Security's full article expands on the conference themes, the evidence gap, and the shift away from checkbox compliance.

Deepen your knowledge

NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, agentic AI identity, machine identity security, and secrets management. It helps practitioners connect identity controls to the continuous risk and audit demands shaping modern security programmes.