AI trust is becoming a continuous GRC control problem

At a glance

What this is: Drata's podcast recap says AI is shifting GRC from periodic assurance to continuous control monitoring, with trust now treated as a live operating state.

Why it matters: This matters because IAM, GRC, and security teams increasingly need evidence, oversight, and approval models that keep pace with AI features, vendor risk, and impersonation threats.

Context

AI trust governance is no longer just about written policy or annual certification. The problem is that AI features, vendor integrations, and impersonation threats can change the effective risk posture faster than traditional review cycles can detect.

For identity and access programmes, that creates a real governance gap. Human identity, non-human identity, and agentic AI controls all need continuous evidence when systems can act, decide, or impersonate at machine speed. A point-in-time control model is now structurally too slow for the risk it is meant to govern.

Key questions

Q: What breaks when AI trust is managed with annual GRC reviews?

A: Annual reviews break down because they assume the control environment stays stable long enough to certify. AI features, vendor changes, and automated decisions can alter risk between review cycles, so documentation can be accurate but operationally outdated. The result is a governance gap where approved controls no longer match current behaviour, especially in workflows that depend on identity, data access, or third-party services.

Q: When should organisations treat a vendor AI feature change as a governance event?

A: Whenever the feature changes data access, decision support, or the approval path for a business process, it should be treated as a governance event. A new AI capability can change the effective risk boundary without changing the vendor name or contract. Security, GRC, and procurement teams should review it before it reaches production use, not after an incident or audit finding.

Q: How do organisations measure whether continuous trust is actually working?

A: Measure the time between a control change and its appearance in governance reporting, the percentage of evidence that is current, and the speed at which exceptions are closed. If those numbers are improving, the programme is becoming more trustworthy. If they are not, the organisation still relies on periodic assurance.

Q: Why are deepfakes a GRC issue as well as a fraud issue?

A: Deepfakes matter to GRC because they can defeat the human checkpoints that many approval and verification workflows still rely on. That makes impersonation a control problem, not just a scam problem. Organisations need stronger identity verification for sensitive actions, and they need governance that assumes voice and video can no longer be trusted on their own.

Technical breakdown

Why point-in-time assurance fails for AI-driven trust

Point-in-time assurance assumes the relevant control state is stable long enough to review, certify, and sign off. AI breaks that assumption because vendor features, workflow automations, and decision logic can change between review cycles. In practice, the control surface shifts from documented policy to live behaviour, including how data is used, which systems are touched, and what actions are taken without fresh human approval. That is why annual evidence packages can still be technically accurate and operationally obsolete at the same time.

Practical implication: move from periodic attestation to continuous evidence collection across AI-linked controls.

Third-party AI risk now sits inside the control perimeter

Third-party AI risk is not limited to vendors that advertise themselves as AI platforms. The article highlights a broader problem: suppliers can embed AI features into existing contracts and services, quietly changing data flows, decision support, and dependency chains. That means the control question is no longer only whether a vendor passed review at onboarding, but whether the current capability set still matches the approved risk boundary. For GRC teams, supplier oversight now has to track feature drift, not just vendor status.

Practical implication: reassess supplier reviews whenever AI functionality is added, changed, or exposed to sensitive workflows.

Trust metrics need to reflect operational behaviour, not paperwork

The article points to measures such as trust-influenced ARR, deal cycle time, and the share of security reviews that close without escalation. Those metrics matter because they connect governance to business flow rather than treating assurance as a back-office compliance exercise. In identity terms, this is a shift from asking whether controls exist to asking whether they actually reduce friction while preserving boundaries. That makes trust measurable, which is the only way it becomes governable at speed.

Practical implication: define live trust indicators that show whether governance is reducing risk without blocking delivery.

Threat narrative

Attacker objective: Exploit stale trust assumptions to gain decision influence, bypass review, or introduce hidden operational and fraud risk.

1. Entry occurs when AI features, vendor changes, or impersonation channels enter the business faster than controls are refreshed.
2. Credentialed or trusted access is then abused because the organisation is still relying on stale assurance, static policy, or weak verification.
3. The impact is governance failure: a trust boundary that no longer reflects how the system actually behaves, decisions that were never fully reviewed, and increased exposure to fraud or vendor-driven risk.

Breaches seen in the wild

• Arup deepfake fraud 2024: Deepfakes of Arup's CFO and colleagues on a video call led a Hong Kong employee to transfer HK$200 million (about US$25.6m) to fraudsters.

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

Continuous assurance is becoming the new control plane for AI governance. Annual review cycles were built for environments that changed slowly enough to document after the fact. AI-driven systems, especially those touching identity, vendor workflows, or automated decisions, now change too quickly for that model to hold. The practical conclusion is that governance must shift from retrospective approval to live evidence and continuous monitoring.

Third-party AI risk now behaves like hidden identity drift. The article shows that vendors can embed AI features into existing services without a clean governance reset. That is an identity problem as much as a procurement problem, because the effective authority, data access, and decision influence of a supplier can change without a new contract event. Practitioners should treat feature drift as a control failure, not a documentation gap.

Deepfakes turn human trust into a security control surface. Once voice and video impersonation can pass routine social verification, traditional identity checks lose much of their defensive value. That means organisations need stronger step-up verification, more resilient approval paths, and explicit trust boundaries for high-risk actions. The conclusion is simple: identity verification must now be designed for adversarial simulation, not just authentication.

GRC is moving from recordkeeping to operational engineering. The article's strongest thread is that trust has to be measured the way performance is measured. That aligns with the emergence of control telemetry, live evidence, and continuously updated assurance. The named concept here is continuous trust operations: governance that tracks current behaviour rather than frozen documentation. Teams that cannot operationalise this will fall behind both AI deployment speed and board expectations.

AI governance now exposes the limits of the annual-audit mindset. Point-in-time compliance still has value, but it cannot be the main defence when AI features, agent behaviour, and third-party dependencies are moving daily. Practitioners should assume that trust gaps will show up first in operational behaviour, not in the next audit cycle. The field is heading toward governance models that are continuously evidenced, continuously scoped, and continuously challenged.

What this signals

Continuous trust operations: AI governance is moving toward live evidence, live boundaries, and live accountability rather than retrospective documentation. That matters for identity teams because the same control logic now has to cover people, vendors, and machine-driven decision paths in real time.

The practical challenge is not whether organisations have policies for AI. It is whether those policies can keep pace with feature drift, impersonation risk, and third-party change without waiting for the next audit cycle.

For practitioners

• Implement continuous control monitoring for AI-linked workflows Track evidence on an ongoing basis for the controls that AI features can alter, including approvals, access paths, and data use. Replace annual-only review points with live checks that show current behaviour, not historical intent.
• Reassess third-party AI risk when vendor capabilities change Trigger a fresh governance review when a supplier adds AI functionality, changes model behaviour, or introduces new decision support into an existing service. Treat feature drift as a new risk event, not a routine update.
• Define trust metrics that leadership can use Use measures such as review cycle time, escalation rate, and the share of security decisions resolved without manual bottlenecks to show whether controls are enabling the business while preserving assurance.
• Harden identity checks against deepfake-enabled impersonation Use stronger verification for high-impact actions, especially where voice or video can be spoofed. Build approval paths that do not rely on a single human-recognition step for payment, access, or policy exceptions.

Key takeaways

• AI has pushed trust out of the audit calendar and into day-to-day control operations, where static evidence is quickly out of date.
• The biggest governance risk is not only model behaviour but supplier and process drift that changes what trusted systems can do without a fresh review.
• Practitioners need continuous evidence, stronger identity verification, and trust metrics that show whether governance is keeping pace with live AI use.

Key terms

• Continuous trust operations: A governance model where trust is monitored, evidenced, and adjusted continuously instead of being certified only at fixed review points. It links policy to live behaviour, so changes in AI features, vendor access, or decision paths are visible before they become incidents or audit findings.
• Trust-influenced metrics: Operational measures that show whether trust controls are supporting the business, not just constraining it. Examples include deal cycle time, escalation rate, and the share of reviews that close cleanly, which help leaders see whether governance is enabling safe movement at speed.
• Scope drift: Scope drift is the gradual mismatch between what an integration was meant to do and what its credentials still allow it to do. It happens when permissions are not revalidated as business needs change, creating hidden over-privilege across SaaS and API-connected systems.
• Deepfake-based impersonation: A fraud technique that uses synthetic audio, video, or both to make an attacker appear to be a trusted person during a live interaction. The tactic exploits human trust in familiar cues and often aims to trigger urgent actions such as payments, resets, or access changes before verification is challenged.

Deepen your knowledge

The NHI Foundation Level course, the industry's only accredited NHI security programme, covers NHI governance, machine identity security, and secrets management in a way that supports modern identity programmes. It helps practitioners connect continuous assurance to the controls their broader security and compliance efforts depend on.