DataSecAI25 shows AI governance is becoming an operational control

At a glance

What this is: Cyera’s DataSecAI25 recap says the field is converging on AI governance as an enabler, with visibility, shadow AI, and agent oversight emerging as the core issues.

Why it matters: For IAM and NHI practitioners, the message is that autonomous agents and unmanaged AI are now governance problems that require identity, access, and data controls to work together.

By the numbers:

• More than 1,000 data, security, AI, privacy, and engineering leaders gathered in Dallas for DataSecAI25.

👉 Read Cyera's DataSecAI25 analysis on AI governance, visibility, and agent oversight

Context

AI governance now sits at the intersection of data security, identity, and access control. The article’s core finding is that leaders no longer view governance as a brake on AI adoption, but as the control layer that makes AI usable at scale without losing visibility into sensitive data or autonomous behaviour.

That matters to IAM and NHI practitioners because the same environment now includes human users, service accounts, AI agents, and unmanaged Shadow AI. The post is less about conference atmosphere than about a structural shift in how organisations will need to govern access, review behaviour, and contain blast radius when AI systems interact with sensitive data.

Key questions

Q: How should security teams govern AI agents that access sensitive data?

A: Security teams should govern AI agents the same way they govern other non-human identities, but with stricter runtime controls. Start with inventory, identity binding, least privilege, and task-scoped access, then add continuous monitoring for unusual behaviour. One-time approval is not enough when an agent can keep acting after onboarding.

Q: Why do AI agents create new risk for IAM and NHI programs?

A: AI agents create risk because they combine execution authority with persistence. Once an agent is permitted to act, it may chain requests, touch data repeatedly, and interact with systems faster than human review can keep up. That makes access scope, monitoring, and revocation speed more important than static policy alone.

Q: What is the difference between shadow AI and approved AI use?

A: Approved AI use sits inside governed workflows with known identities, policies, and monitoring. Shadow AI operates outside those controls, often through personal accounts or unsanctioned tools, which means the organisation may not know what data is exposed or which permissions are in play. The governance gap is visibility, not just policy.

Q: When should organisations move from manual review to automated AI governance?

A: Organisations should move to automated governance when AI systems begin making repeated decisions, requesting frequent access, or touching sensitive data at scale. Manual review works for low-volume experimentation, but it breaks down once agent activity becomes continuous. Automation becomes necessary when delay itself becomes a security risk.

Background and context

Why unified visibility is now the control plane for AI governance

Unified visibility means correlating data location, sensitivity, identity, and access into one control picture. In AI-heavy environments, point tools can show a secrets issue, a data exposure, or an identity anomaly, but rarely the full chain from source data to the actor touching it. That is why leaders at the event kept returning to the need to know what data exists, where it lives, and who or what can reach it. For IAM and NHI teams, the technical point is simple: without joined-up visibility, policy decisions are always late and partial.

Practical implication: Build access review and data discovery processes around shared identity and data telemetry, not isolated tool outputs.

How shadow AI changes the trust model for non-human identities

Shadow AI is unmanaged AI usage that appears outside approved controls, often through SaaS tools, personal accounts, or unreviewed workflows. Unlike traditional applications, AI systems can generate requests, chain actions, and move between systems with little human friction. That creates a trust problem for NHI governance because the organisation may not know which agent exists, which identity it uses, or which permissions it has inherited. The result is hidden execution authority rather than merely hidden software.

Practical implication: Treat every unmanaged AI workflow as an unreviewed non-human identity until it is inventoried, authorised, and constrained.

Agent oversight requires runtime controls, not just approval workflows

AI agents change the operational model because they request access continuously and act autonomously once permitted. Manual approval gates may still matter at onboarding, but they do not address the steady stream of decisions made during execution. The article’s example of an agent logic loop driving cloud costs into the millions shows the failure mode clearly: a permitted agent can become an expensive or risky actor long after initial access is granted. Runtime governance therefore matters more than one-time approval.

Practical implication: Use runtime policies, anomaly detection, and task-scoped access to limit what agents can do after they are approved.

NHI Mgmt Group analysis

AI governance is becoming an identity problem before it is a data problem. Once humans, bots, and agents all interact with sensitive datasets, access control becomes the operational backbone of governance. The article reflects a broader market shift: security teams are being asked to govern execution, not just data storage. Practitioners should design governance programmes around identity-led enforcement.

Shadow AI creates a hidden NHI population that traditional asset inventories will miss. Unapproved AI use is not just a policy violation, it is a control gap because the organisation may never map the identities, tokens, and permissions involved. That leaves access review blind to the real actors touching data. Practitioners should assume unmanaged AI already exists and build detection for it.

Runtime supervision is the new boundary for AI risk. A one-time approval model does not work when autonomous systems keep acting after initial access is granted. The meaningful control question is whether the organisation can observe, constrain, and stop agent behaviour in real time. Practitioners should move from static governance to continuous enforcement.

Data visibility and NHI governance are converging into one programme. Teams that separate data security from identity governance will miss the combined blast radius created when an over-permissioned agent can reach sensitive content. That convergence is where policy, access review, and anomaly detection need to meet. Practitioners should treat AI governance as a shared operating model across security, data, and IAM.

Orchestration is becoming the right operating metaphor for security teams. The post signals a move away from siloed control ownership toward coordinated enforcement across data, access, and AI workflows. That does not mean centralising every decision, but it does mean standardising the controls that govern machine behaviour. Practitioners should prepare for cross-functional governance boards with technical enforcement behind them.

From our research:

• Only 1.5 out of 10 organisations are highly confident in their ability to secure NHIs, compared to nearly 1 in 4 for securing human identities, according to The State of Non-Human Identity Security.
• Only 85% of organisations lack full visibility into third-party vendors connected via OAuth apps, which shows how quickly access sprawl can outpace review processes.
• The next step is to pair identity visibility with governance action, as outlined in Ultimate Guide to NHIs , Key Challenges and Risks.

What this signals

Identity-led governance is now the practical path for AI programmes. The organisations that can correlate data sensitivity, identity state, and runtime behaviour will move faster with less guesswork. That means IAM teams should plan for AI oversight as an operational capability, not a policy exercise. The shift is already underway, and programmes that stay siloed will struggle to keep pace.

Ephemeral access is not the same as governed access. If an agent can continuously request new actions, the real issue is not whether credentials expire, but whether the organisation can observe and constrain behaviour in time. Teams should expect more demand for runtime enforcement, access revocation automation, and alerting tied to actual data movement rather than entitlement counts.

The governance model will increasingly resemble orchestrated control across data, identity, and AI execution. Teams that anchor their programme in NIST Cybersecurity Framework 2.0 can map the shift from visibility to protection to response without forcing AI into a human-only model. That framework fit matters because autonomous actors need continuous control, not periodic review.

For practitioners

• Inventory shadow AI workflows Map every approved and unapproved AI workflow that can touch sensitive data, including SaaS tools, personal accounts, and agentic automations. Classify the identities, tokens, and permissions each workflow uses before allowing continued access.
• Tie data discovery to access review Run access reviews against live data sensitivity and usage telemetry so reviewers can see which human and non-human identities actually reach sensitive stores. Use the combined view to remove stale access and reduce hidden blast radius.
• Apply runtime controls to agent activity Set task-scoped limits, anomaly detection, and kill-switch procedures for autonomous agents that can act continuously after approval. Focus on cost spikes, unusual request loops, and unexpected data movement.
• Separate experimentation from production authority Allow AI experimentation in constrained environments, but require explicit promotion before any workflow can access production data or systems. Keep the policy boundary clear so curiosity does not become standing access.

Key takeaways

• AI governance is moving from a slowing mechanism to a control layer that makes AI adoption safer and more scalable.
• The hardest part of the problem is visibility across data, human users, and non-human identities, not policy language.
• Security teams should prepare for continuous runtime oversight of AI agents instead of relying on one-time approval gates.

Key terms

• Shadow AI: Shadow AI is AI use that happens outside approved governance, monitoring, or procurement channels. In practice it often appears through personal accounts, unsanctioned tools, or unreviewed automations, creating hidden data exposure and identity risk that security teams cannot see in normal inventories.
• AI Agent: An AI agent is an autonomous software entity that can execute actions, call tools, and make operational decisions within assigned limits. For security teams, it should be treated as a non-human identity because it can hold permissions, interact with systems, and create risk without direct human intervention.
• Runtime governance: Runtime governance is the control of behaviour while software is actively executing, not just before it is approved. In AI environments it includes monitoring, policy enforcement, and interruption of abnormal actions so that access decisions remain aligned with actual system behaviour.

Deepen your knowledge

AI governance, non-human identity inventory, and runtime access control are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If your programme is already dealing with shadow AI or autonomous agents, the course provides a practical starting point.

This post draws on content published by Cyera: What DataSecAI 2025 Revealed About the Future of Data, Security, and AI. Read the original.