AI adoption is turning governance into a real-time identity problem

At a glance

What this is: This is a Nudge Security research report showing that AI adoption has become operational, with governance now driven by real-time tool usage, integrations, and data exposure rather than policy documents alone.

Why it matters: It matters because IAM, NHI, and emerging agentic AI controls must account for how people, permissions, and platforms interact in day-to-day workflows, not just what was approved at onboarding.

By the numbers:

• OpenAI is present in 96.0% of organizations, according to Nudge Security.
• Anthropic is present in 77.8% of organizations, according to Nudge Security.
• 17% of prompts include copy-paste or file upload activity, according to Nudge Security.
• Sensitive-data events are led by secrets and credentials at 47.9%, according to Nudge Security.

Context

AI governance is no longer just about approving tools before they reach users. In practice, employees are now using AI inside business workflows, connecting it to other systems, and sharing data through prompts and files, which turns governance into an identity and access problem that changes continuously.

That shift matters for both human IAM and NHI governance. When AI tools connect to productivity suites, repositories, and knowledge systems, the control question is no longer whether a tool is allowed, but what it can reach, what it can send, and how quickly those permissions can change.

Key questions

Q: How should security teams govern AI-enabled workflows that can act on their own?

A: Treat them as identity-governed execution paths, not just software features. Assign a named owner, define least-privilege access, log every tool call, and require revocation paths for credentials and tokens. If the workflow can touch production systems or sensitive data, its permissions must be reviewed with the same discipline used for privileged machine identities.

Q: Why do one-off AI integrations create risk in identity security programmes?

A: They create many separate control paths, each with its own testing, maintenance, and logging burden. Over time, those paths drift apart, so authentication context, policy checks, and audit evidence become inconsistent across workflows that should be governed the same way.

Q: What are the signs that AI governance is failing in the enterprise?

A: Common warning signs include rapid growth in AI use without matching policy coverage, sensitive files being copied into personal accounts, and a large share of AI apps carrying high or critical risk. Another indicator is weak visibility into who is using which tools and what data they are sending. If teams cannot answer those questions, governance is not working as intended.

Q: Should organisations prioritise continuous monitoring or periodic access reviews for audit readiness?

A: Continuous monitoring should come first when environments change quickly or privileged access spans multiple systems. Periodic access reviews still matter, but they are too slow to prove ongoing control in dynamic estates. The best sequence is live visibility first, then scheduled review for governance confirmation and exception handling.

Technical breakdown

Why embedded AI changes the governance model

Embedded AI changes the governance model because the risk is created by runtime behaviour, not by a one-time approval decision. Once AI tools sit inside collaboration, coding, and knowledge workflows, they can move data across trust boundaries without looking like a traditional application. That makes static policy reviews incomplete. The governance problem becomes one of observing tool usage, integration scope, and data egress as they happen, then tying those observations back to identity, ownership, and permitted access paths.

Practical implication: shift governance from point-in-time approvals to continuous discovery of where AI is active and what it can touch.

How AI integrations create identity and access exposure

AI integrations matter because the AI service inherits practical access through the systems it connects to, even when the user thinks they are only using a chat tool. In enterprise environments, that means productivity suites, code repositories, and knowledge systems can become part of the AI attack surface. For IAM teams, the technical issue is delegated access and downstream propagation of permissions. For NHI teams, the parallel issue is whether service-to-service and app-to-app trust has been expanded without explicit governance.

Practical implication: inventory every AI integration as an access path, not just as a productivity feature.

Why prompts now function like a data egress channel

Prompts now behave like a lightweight exfiltration path because users can paste content or attach files into AI tools with little friction. That does not make every prompt malicious, but it does mean sensitive data can leave approved control boundaries in ways that ordinary DLP and approval workflows may not see soon enough. Secrets, credentials, financial details, and health data are especially exposed because they are easy to paste and hard to govern after the fact. The control challenge is to treat prompt content as governed data movement.

Practical implication: align prompt monitoring and DLP with the same sensitivity rules used for other outbound data flows.

Breaches seen in the wild

• CoPhish OAuth phishing via Copilot Studio: Datadog showed Copilot Studio agents on a Microsoft domain can front OAuth consent phishing and forward stolen tokens; no victims reported.
• AI agent retail card theft campaign 2026: AI agents breached 27+ retailers for about $25 each, used cloud keys and a Secrets Manager dump, and stole 600,000+ payment cards.

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

AI governance has become an identity governance problem, not a policy problem. The article shows that AI is now embedded in workflows and connected to operational systems, which means the real control surface is who or what can act, connect, and transmit data at runtime. Policy still matters, but policy alone cannot govern live permission paths. Practitioners should treat AI usage as a governed identity state, not a static acceptable-use category.

Real-time visibility is the missing governance primitive for AI adoption. Approval logs and annual reviews cannot answer what an AI tool is connected to today, which data it touched this morning, or whether its scope expanded through a new integration. That is why continuous discovery matters more than periodic certification in this category. The operational conclusion is that governance must follow usage, not just procurement.

Secret exposure is now the most important AI data-loss signal. When secrets and credentials dominate sensitive-data events, the issue is not merely user misuse but the ease with which privileged material enters prompts and attachments. That creates a direct overlap between AI governance, NHI hygiene, and incident response. The practitioner takeaway is to unify AI monitoring with credential protection rather than running them as separate programmes.

AI integrations expand the blast radius of existing trust relationships. The article makes clear that AI tools increasingly plug into productivity, code, and knowledge systems, which means each connection inherits upstream and downstream risk. This is where identity teams need to think in terms of scoped trust, delegated access, and revocation paths. The practical implication is that integration review is now a core part of AI governance.

Prompt behaviour is becoming a named governance concept in its own right. The article points to a prompt egress pattern, where copy-paste and file upload activity become the mechanism by which sensitive data leaves control boundaries. That concept matters because it links human behaviour, application integration, and information exposure in one operational model. Practitioners should classify prompt flows as governed data movement, not informal user interaction.

From our research library:

• Organisations that describe themselves as confident in their AI deployment actually experience a 72% security incident rate, compared to 33% for those who remain cautious, according to the 2026 Infrastructure Identity Survey.
• 19% of organisations give AI systems dramatically more access than human employees, nearly one in five granting unrestricted privilege, according to the 2026 Infrastructure Identity Survey.
• Read next: AI Agent Observability, Audit and Incident Response Guide

What this signals

Real-time AI governance is becoming the default operating model: approval gates and acceptable-use policies do not tell teams which AI systems are active, what they are connected to, or whether they are handling sensitive data right now. That forces identity teams to govern usage, not just authorization.

The practical shift for IAM and NHI leaders is to treat AI integrations as first-class access paths. Once a tool can reach a repository, productivity suite, or knowledge platform, it belongs in the same control conversation as any other privileged connection.

For security programmes, the biggest mistake is separating AI oversight from data protection and identity lifecycle work. The article points to a joined-up problem where human behaviour, delegated access, and prompt-based data movement need one operating model.

For practitioners

• Map AI tools to live access paths Inventory which AI tools are connected to productivity suites, knowledge systems, and code repositories, then document the permissions each integration can exercise. Treat every integration as a governed identity path rather than a standalone app choice.
• Classify prompt content as data movement Apply sensitivity rules to copy-paste and file upload activity inside AI tools so that secrets, credentials, and regulated data are monitored as outbound flows. Align those rules with existing DLP and incident-response playbooks.
• Review delegated access behind AI integrations Check whether AI-connected services inherit broader permissions than the business use case requires, especially in chat, coding, and knowledge workflows. Revoke or narrow access where the integration can reach more systems than the user intended.
• Build continuous AI usage visibility Replace one-time approval checkpoints with ongoing discovery of where AI is active, what systems it reaches, and whether new integrations have appeared without governance review. Use that telemetry as the basis for recertification and exception handling.

Key takeaways

• AI adoption has crossed into operational workflows, which makes governance a live identity problem instead of a policy-only problem.
• The strongest evidence in the report is the combination of broad AI usage, widespread integrations, and prompt-driven sensitive data exposure.
• Security teams need continuous visibility into AI connections, prompt activity, and delegated access if they want governance to keep pace.

Key terms

• AI Governance: AI governance is the set of controls used to discover, classify, approve, restrict, monitor, and revoke AI-enabled access. It connects identity, data, and policy so organisations can manage what AI can reach, what it can share, and when it should be stopped.
• Prompt Egress: The movement of sensitive information out of a controlled system through prompts, uploads, or pasted content. In AI governance, this is a data-loss pathway because the input channel can bypass the original application’s access model and expose secrets or regulated data elsewhere.
• Delegated Access: Delegated access is permission granted to one identity to act on behalf of another user, service, or system. In NHI environments, this usually appears in OAuth-connected apps and automation tooling. It is powerful, but it must be tightly scoped and reviewed because it can persist long after the original business need ends.
• Runtime Visibility: The ability to observe what an AI client actually accessed, which tools it used, and how it behaved during a session. It is more useful than entitlement snapshots for agent governance because it captures executed reality, not just approved access.

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.