AI adoption is outpacing governance in professional services

At a glance

What this is: Cyberhaven’s report says professional services AI adoption is accelerating faster than governance, with usage spreading across GenAI SaaS, endpoint AI applications and AI agents.

Why it matters: This matters because identity, access and data controls lose effectiveness when employees use multiple AI tools and unmanaged environments outside normal oversight.

Context

Professional services firms are adopting AI across sanctioned apps, endpoint tools and agentic workflows faster than their governance models can track. The core problem is not AI enthusiasm on its own, but fragmented usage across accounts, devices and environments that weakens oversight, data control and accountability.

When AI activity spreads beyond traditional IT visibility, the control plane shifts from policy documents to day-to-day identity, data and access enforcement. That creates a governance gap for human users, service accounts and AI agents alike, because the organisation can no longer assume it knows where sensitive data is going or which environment is in scope.

Key questions

Q: What breaks when AI adoption outpaces governance?

A: What breaks first is attribution. Teams lose visibility into which tools are in use, which data they can reach, and which actions were taken automatically versus manually. Once adoption is ahead of control design, security teams end up retrofitting policy around live workflows instead of governing them from the start.

Q: Why does fragmented AI infrastructure create security risk?

A: Fragmented AI infrastructure creates risk because each provider handoff can split responsibility for credentials, permissions, and logging. When teams stitch together GPU services, model hosting, and cloud platforms, the result is often inconsistent governance and hidden access paths. That makes it harder to prove who can reach what and under which conditions.

Q: How do security teams know if AI governance is working?

A: Look for evidence that access decisions are reviewable, permissions are revocable, and exceptions are not becoming permanent. If the team cannot explain who owns an AI workflow, what it can reach, and when its access was last reviewed, governance is incomplete. Control maturity shows up in traceability, not adoption volume.

Q: How should organizations approach the governance of AI agents?

A: Organizations should adopt a governance framework that incorporates continuous visibility, adaptive IAM practices, and stringent policy-based controls. This ensures that all agent actions are tracked, authorized appropriately, and assessed for compliance.

Technical breakdown

Why fragmented AI usage breaks governance

AI governance assumes a manageable set of approved tools, known users and stable access paths. In practice, professional services adoption is splitting across GenAI SaaS, endpoint applications, personal accounts and unmanaged environments. That fragmentation makes inventory, policy enforcement and data classification harder because the same worker may interact with multiple AI surfaces in one workflow. The technical problem is not just tool sprawl, but control-plane sprawl: visibility, logging and enforcement are no longer anchored to a single enterprise-managed path.

Practical implication: governance must follow the actual usage path, not the approved-tool list.

How data exposure increases across AI endpoints and agents

AI systems become risky when they can observe, transform or forward sensitive content without sufficient boundary controls. Endpoint AI applications and AI agents are especially problematic because they sit close to user activity and can operate across contexts, which makes data loss prevention, content control and session-level monitoring harder to apply consistently. In this setting, the issue is not that AI always exfiltrates data, but that the enterprise often cannot see whether sensitive material was copied into prompts, outputs or downstream tasks before it left controlled systems.

Practical implication: control sensitive-data flow at collection, prompting and handoff points, not only at storage.

Why AI agents need identity-aware governance

An AI agent is a software entity that can select actions and tools at runtime, so it behaves more like an active workload than a static application. That means its identity, permissions and context matter as much as the model it uses. If agent access is not tied to lifecycle governance, an organisation can end up with agents acting through personal accounts, over-broad tokens or ambiguous ownership. The governance failure is not agent autonomy by itself, but the absence of a reliable identity model that defines who authorised the agent, what it may touch and when its access ends.

Practical implication: treat agent identity, privilege and lifecycle as first-class controls.

Threat narrative

Attacker objective: The practical objective is not always a classic intrusion, but uncontrolled access to sensitive business data through AI-enabled workflows that the organisation cannot adequately monitor or govern.

1. Entry occurs when employees introduce AI tools through sanctioned apps, endpoint software or unmanaged personal environments, expanding the number of places where sensitive work data can be exposed.
2. Credential and context abuse follows when accounts, tokens or connected sessions let those tools operate across multiple environments without consistent oversight.
3. Impact emerges as sensitive information moves through prompts, outputs and downstream workflows outside the visibility of traditional IT control planes.

Breaches seen in the wild

• McKinsey AI platform breach: McKinsey AI platform hack exposed 46M chats and sensitive data.
• DeepSeek database exposure 2025: An unauthenticated DeepSeek ClickHouse database exposed over a million log lines with plaintext chat history and API keys in 2025.

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 adoption governance is now a control-plane problem, not a policy problem. Once employees distribute AI usage across SaaS apps, endpoint tools and unmanaged environments, the enterprise loses the ability to enforce a single governance model. That breaks the assumption that one approved catalog or one identity boundary can describe actual AI use. Practitioners should treat governance as an operating-state issue, not a document-review exercise.

Professional services has a distinct AI exposure profile because knowledge work is data-dense and session-heavy. The combination of client documents, drafts, summaries and rapid tool switching creates many opportunities for sensitive data to cross boundaries invisibly. This is where identity governance and data governance intersect, because the question is not only who can use AI, but what that user or agent can see, copy and forward. Practitioners need controls that bind usage to context, not just to login status.

AI agents sharpen the governance debt already present in fragmented AI adoption. The moment a software entity can act across tools, the old assumption that a human session is the primary unit of accountability becomes weaker. That is why agent identity, token scope and lifecycle ownership belong in the same conversation as AI policy. Practitioners should assume that unmanaged agent privileges will create hidden access paths unless the identity model is explicit.

Data exposure is the consequence that converts AI governance from a future concern into an immediate security issue. The report’s central message is that the organisations moving fastest are also the ones most likely to outrun their own oversight. In that environment, the security outcome is not simply more AI use, but more uncontrolled movement of sensitive information. Practitioners should reframe AI governance as exposure reduction across identity, endpoints and workflows.

Governance fragmentation is the named concept this report exposes. It describes the gap that appears when AI adoption, access control and data oversight are managed as separate programmes instead of one operating model. The practical lesson is that AI governance cannot be effective if the organisation cannot see all AI paths, all identities and all data handoffs. Practitioners should collapse those silos before adoption grows further.

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.
• Only 44% of organisations have implemented any policies to manage their AI agents, despite 92% agreeing that governing AI agents is critical to enterprise security, according to the 2026 Infrastructure Identity Survey.
• Read next: Agentic AI Identity Risk Board Briefing

What this signals

Governance lag becomes visible first in data movement, not in policy language. When employees route work through GenAI SaaS, endpoint AI and unmanaged accounts, the organisation can still sound governed while losing practical control. The reader-facing signal is to measure where data actually flows, because AI governance only matters once it changes day-to-day behaviour.

AI adoption now behaves like an identity and access problem with a data-loss outcome. The same control gaps that have long affected shadow IT now appear in AI workflows, but with faster iteration and more sensitive content. Security leaders should expect the next governance failure to come from uncontrolled sessions, not from a single sanctioned platform.

Policy without runtime enforcement creates governance debt. Professional services teams need to prepare for a world where approvals do not guarantee containment. The practical implication is to align AI policy, identity lifecycle and data handling in one operating model before adoption grows further.

For practitioners

• Map all AI usage paths Inventory sanctioned GenAI SaaS, endpoint AI applications, personal accounts and agent workflows so governance reflects real usage rather than approved-tool assumptions.
• Bind AI access to identity and lifecycle controls Define ownership, approval, privilege scope and offboarding for human users, service accounts and AI agents that touch sensitive work data.
• Control sensitive data at prompt and handoff points Apply data handling rules where content is entered into AI systems, transformed by them or passed into downstream tasks, not only where it is stored.
• Separate unmanaged experimentation from governed production Create explicit boundaries between exploratory AI use and business workflows that handle client or regulated information, then monitor crossings between the two.
• Review agent token scope and session duration Limit what AI agents can access, ensure credentials expire with the task and verify that no agent keeps broader access than its current job requires.

Key takeaways

• Professional services AI adoption is spreading across multiple tool types and access paths faster than governance can consistently track.
• The main risk is now uncontrolled data movement through approved and unmanaged AI workflows, not isolated experimentation.
• Security teams need governance that follows identity, session and data handoff points, because policy alone no longer defines the control boundary.

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.
• 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.
• Control-plane sprawl: Control-plane sprawl is the condition where identity, device, privileged access, and provisioning are managed in separate systems with weak policy coordination. It usually creates duplicated administration, inconsistent enforcement, and slower offboarding because no single workflow owns the whole lifecycle.
• Sensitive Data Exposure: Sensitive data exposure is the condition where high-value or regulated information is reachable by identities, applications, or services beyond its intended scope. In modern environments, exposure is often driven by sprawl, duplicated storage, and poor entitlement visibility rather than a single leaked file.

Deepen your knowledge

NHI Mgmt Group covers identity security, NHI governance, and agentic AI through independent research, practitioner guides, and the NHI Foundation Level course. It is suited to practitioners who need a common language for governing non-human identities, access scope and lifecycle control.