Shadow AI governance is outpacing enterprise access controls

At a glance

What this is: This is Delinea's analysis of shadow AI, showing that unsanctioned AI use is already linked to breaches, compliance issues, and data risk in mid-sized firms.

Why it matters: It matters because IAM, IGA, and PAM teams now have to govern AI access and data use patterns that sit outside approved identity workflows, especially where privileged data is involved.

By the numbers:

• Shadow AI is rampant in mid-sized firms, with 81% of organisations in the $10 million to $50 million revenue range facing breaches or compliance issues due to unauthorized AI use.

Context

Shadow AI is the unsanctioned use of AI tools by employees or departments outside IT or security oversight. In identity terms, the problem is not just tool sprawl, but access sprawl: unknown AI systems can read, transform, retain, or exfiltrate data without the governance controls that normally define who or what may touch it.

Delinea frames this as a governance lag problem, and the evidence suggests that lag is already creating operational exposure. For IAM and PAM teams, the relevant question is no longer whether AI is being used, but which identities, datasets, and workflows it is quietly sitting inside.

The article's examples include developers using LLM prompts for code generation and finance teams using unauthorized AI agents for payroll-related work. That makes the subject broader than shadow IT, because the access path is the risk: if the tool is unsanctioned, the identity behaviour and downstream data handling are unsanctioned too.

Key questions

Q: What breaks when shadow AI is not mapped in the access graph?

A: Teams lose visibility into which service accounts, tokens, or delegated credentials give hidden AI systems access to tools and data. Without that mapping, least privilege cannot be measured properly, and offboarding may miss the real execution path that keeps the agent alive.

Q: Why does shadow AI create such a high breach risk?

A: Shadow AI creates high breach risk because it can access sensitive data outside normal oversight, then process or reproduce that data in places security teams do not control. Once the tool can read privileged information, the organisation may lose visibility into retention, logging, and downstream exposure.

Q: How do you know if shadow AI governance is actually working?

A: You know it is working when you can see where AI is used, what data it touches, what actions it can take, and whether those actions are blocked or approved in real time. If usage is still being discovered through incidents or audits, governance is lagging behind adoption.

Q: What breaks when organisations ban shadow AI instead of governing it?

A: Bans often push AI use into personal accounts, unmanaged devices, and hidden workflows, which removes visibility from security and makes data exposure harder to detect. The control failure is not usage itself, but concealment. A better model is to approve fast, workable alternatives and enforce policy on identity, data handling, and logging.

Technical breakdown

Why shadow AI creates an access governance problem

Shadow AI becomes an identity issue when an unsanctioned model, assistant, or agent is allowed to operate against enterprise data without being enrolled in the organisation's access model. The control gap is not only inventory. It is the lack of a defined identity, approved scope, and auditable relationship between the AI tool and the data it can reach. Once the tool can query, summarize, or transmit information, traditional approval boundaries stop being meaningful unless access is explicitly bound to a governed identity and data classification model.

Practical implication: Treat unsanctioned AI use as an access governance failure, not just an application risk.

Why data visibility matters more than tool visibility

The article's core warning is that once an AI tool touches data, teams often lose track of where that data goes next. That makes data visibility the decisive control plane. If security teams cannot see which datasets a model touched, whether prompts contained sensitive content, or whether outputs were retained elsewhere, they cannot separate benign productivity from exposure. This is especially important for privileged data, where the business impact of leakage is higher and the normal assumptions around request, approval, and retention no longer hold.

Practical implication: Map AI use to the sensitive data it can access and monitor the resulting data paths, not just the presence of the tool.

How governance lag turns AI adoption into shadow AI

Governance lag is the mismatch between how quickly users adopt AI tools and how slowly policy, approval, and enforcement catch up. In practice, that lag produces a control blind spot: people use tools before the organisation has decided whether those tools are allowed, monitored, or constrained. The article also points to a common failure pattern in which teams focus on banning tools instead of defining acceptable use, access scope, and escalation paths. That usually pushes usage further underground rather than shrinking the risk surface.

Practical implication: Build policy and controls around actual user behaviour so governance arrives before adoption becomes entrenched.

Threat narrative

Attacker objective: To obtain or move sensitive enterprise data through an unmanaged AI channel that bypasses approved access controls.

1. Entry occurs when employees or departments adopt unsanctioned AI tools outside IT or security oversight, often through ordinary productivity workflows.
2. Credentialed or privileged data is then exposed to the tool, which can retain, transform, or relay information beyond the organisation's approved boundaries.
3. The organisation loses visibility into where the data ends up, creating escalation into compliance failure, breach exposure, or broader exfiltration risk.
4. Impact follows when sensitive data, code, or operational information is accessed through an unmanaged AI path that governance cannot reliably reconstruct.

Breaches seen in the wild

• Samsung ChatGPT leak 2023: Samsung staff pasted chip source code and meeting notes into ChatGPT weeks after it was allowed, leading Samsung to restrict generative AI tools.
• OmniGPT breach claim 2025: A hacker claims to have leaked 34 million OmniGPT AI chat messages holding users' API keys and credentials; OmniGPT has not confirmed it.

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

Shadow AI is first and foremost an access governance problem. The article is not describing a simple policy nuisance. It is showing that AI tools become material when they are allowed to touch enterprise data without a stable identity, scope, or lifecycle boundary. That means IAM, IGA, and PAM teams have to treat shadow AI as a governed access path, not as a side issue for the security awareness team.

Data visibility is the control that separates productivity from exposure. Delinea's framing is correct that the biggest risk is what the AI can see and where that data can go next. Once prompts, outputs, or embedded datasets become opaque, classic approval-based governance loses traceability. Practitioners need to think in terms of data reach and retention, not just application inventory.

Shadow AI governance lags because organisations still rely on sanctioned-workflow assumptions. That assumption works when access is centrally approved and observable. It breaks when users can introduce AI tools into business processes faster than policy can classify them, which is why governance has to shift from prohibition to controlled scope and data-bound authorization.

AI access-to-data blast radius: the real risk is not the presence of AI, but the size of the data footprint an unsanctioned tool can touch before governance notices. The wider that footprint, the faster a productivity choice becomes a breach condition. This is the concept security leaders should use when they explain shadow AI to the board.

Mid-sized firms are the warning segment, not an edge case. The article's 81% figure shows that governance debt is not confined to highly complex enterprises. It appears wherever adoption outpaces controls, which makes shadow AI a mainstream identity governance issue rather than a niche AI security topic.

From our research library:

• AI-related credential leaks surged 81.5% year-over-year in 2025, with the surrounding AI infrastructure leaking 5x faster than core LLM providers, according to the State of Secrets Sprawl 2026.
• Read next: AI Agent Authorisation Guide

What this signals

Shadow AI governance has become a data-access issue, not just an acceptable-use issue. Security teams should assume that unsanctioned AI will continue to appear inside ordinary business workflows, which means the practical control point is data access, not tool approval. If an AI system can reach privileged or regulated information without a formal owner, the organisation has already lost part of the control plane.

AI access-to-data blast radius: the metric that matters is not how many AI tools are present, but how much sensitive information each one can touch before detection. That shift in thinking aligns shadow AI with identity governance, because the question becomes which identity is allowed to expose which data under which conditions.

This topic also points to a programme design change: AI policy, access management, and data classification can no longer be maintained as separate workstreams. They have to move together, or users will keep finding the gap between authorised technology and unauthorised behaviour.

For practitioners

• Inventory unsanctioned AI touchpoints Identify where employees are already using LLMs, copilots, and AI agents in business workflows, then classify those touchpoints by data sensitivity and business function.
• Bind AI use to data classification Set access rules that limit what sensitive datasets AI tools can reach, with stricter handling for privileged data and regulated information.
• Monitor prompt and output paths Track where prompts, generated content, and exported outputs flow so teams can detect when sensitive information leaves approved boundaries.
• Define approved AI use cases Publish clear use-case boundaries for AI adoption so users know which tools are allowed, what data they may process, and when escalation is required.
• Review access for hidden AI identities Look for AI tools acting like unmanaged identities in internal processes, especially where they can read or move data without a formal owner.

Key takeaways

• Shadow AI is already producing real governance failures, including breaches and compliance issues, rather than remaining an emerging concern.
• The core exposure is uncontrolled access to sensitive data through unsanctioned AI tools, not AI adoption on its own.
• IAM, PAM, and data governance teams need shared controls that make AI use visible, bounded, and tied to approved data access.

Key terms

• Shadow AI: AI agents, copilots, or connected tools operating without full visibility or governance from security teams. Shadow AI becomes an identity problem when those systems authenticate with unmanaged tokens, service accounts, or OAuth apps that can reach production resources.
• AI Access Event Governance: AI access event governance is the practice of treating every meaningful AI tool action as part of the identity and audit model. It links access, lifecycle, and evidence so that AI usage is governed as an enterprise control surface rather than an informal productivity layer.
• Governance Latency: Governance latency is the delay between a change in risk, relationship, or access need and the point at which the control model reflects that change. In API environments, high governance latency turns simple access management into a bottleneck and increases residual exposure.
• Data Access Boundary: A data access boundary is the practical limit around which information a tool or identity may read, transform, or export. For AI use, it matters because the main risk is not the model itself, but the sensitive data it can touch once users place it in the workflow.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity lifecycle are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an IAM programme, it is worth exploring.