By NHI Mgmt Group Editorial TeamBased on Akto: “Shadow AI Risks in Enterprises: Detection, Governance & Security” (May 29, 2026)

TL;DR: Shadow AI is spreading through personal accounts, unapproved browser extensions, API integrations, and autonomous agents that can access enterprise data and systems outside approved governance, according to Akto. The core issue is inventory and control gap, not just data leakage, because AI actions can hide inside normal user activity and bypass traditional discovery.

At a glance

What this is: Akto argues that shadow AI becomes an enterprise governance failure when unapproved tools, personal accounts, browser extensions, and autonomous agents operate outside visibility and control.

Why it matters: IAM, security, and AI governance teams need to treat shadow AI as an access and lifecycle problem, because hidden AI activity can bypass approval, logging, and least-privilege controls.

By the numbers:

  • More than 80% of workers use unapproved AI tools, according to Akto.
  • Only 37% of organisations have a documented AI governance policy, according to Akto.
  • Around 69% of organisations either do not monitor AI usage or address it only reactively, according to Akto.

Context

Shadow AI is not just an application-safety issue. It is a governance gap created when AI tools, assistants, extensions, and agents are adopted faster than enterprises can inventory, approve, and monitor them. In practice, that makes the control problem an identity and access problem as much as a data-handling problem, because many of these tools operate through user accounts, OAuth grants, API keys, or embedded permissions.

The operational failure is familiar to IAM teams: if a system is not in the inventory, it is outside lifecycle control. That means no reliable ownership, no consistent approval path, no clean offboarding, and no trustworthy audit trail for AI-mediated actions. The article’s examples show that shadow AI is often less visible than shadow IT because it can sit inside ordinary browser activity and approved SaaS workflows.

Key questions

Q: Where does shadow AI governance fail first in enterprises?

A: It usually fails first at discovery. If security teams cannot inventory AI tools, agents, extensions, and embedded copilots, they cannot approve them, assign owners, classify the data they touch, or retire them cleanly when risk changes. Governance starts with visibility, because unseen tools are outside every later control.

Q: Why is shadow AI harder to manage than ordinary shadow IT?

A: Shadow AI is harder because the tools can process sensitive content as part of normal use, which creates both data exposure and compliance risk. Traditional SaaS oversight often tracks application presence, but AI governance must also account for how data is entered, stored, and audited.

Q: What are the signs that shadow AI agents are escaping discovery controls?

A: Common signs include identity traces without formal registration, endpoint-level activity from local agents or MCP servers, and broad permissions with no clear owner. When those signals show up across SaaS, IDP, and EDR data but never appear in the central catalogue, discovery is failing. The symptom is fragmentation between where the agent acts and where governance can see it.

Q: Should organisations treat AI pentesting agents like non-human identities?

A: Yes. If an agent can authenticate, browse, test, and interact with systems, it has an authority boundary that should be governed like any other non-human identity. That means ownership, least privilege, explicit scope, isolation, and revocation controls. The difference is that these controls must be runtime-enforced, not assumed from policy.

Technical breakdown

Why shadow AI is harder to inventory than shadow IT

Shadow AI differs from classic shadow IT because the workload is often hidden inside legitimate channels such as browser sessions, consumer AI accounts, SaaS copilot features, or extensions with broad page access. The system may look like normal user activity while actually sending prompts, files, or business context to a third-party model. That makes discovery harder than endpoint or SaaS inventory alone. For security teams, the important architectural point is that the control boundary has shifted from a visible application to a mix of identity, browser, API, and workflow layers.

Practical implication: Treat AI discovery as a cross-layer control problem, not a single-tool scan.

How autonomous agents change the access model

Autonomous AI agents are materially different from chat interfaces because they can take actions, not just generate content. When an agent uses OAuth tokens, stored credentials, or API keys, it can query systems, send mail, write files, and trigger workflows without a human at each step. That creates a machine-identity-like access pattern even when the agent is launched by a person. The core governance challenge is that the action trail may be attributed to the human account while the decision logic comes from the agent, which weakens traditional accountability and review models.

Practical implication: Require explicit identity ownership and scoped permissions for every agent that can act on enterprise systems.

Why approved SaaS can still become a shadow AI path

Shadow AI does not always enter through unsanctioned tools. Many approved SaaS products now include embedded AI features that process customer data, meeting notes, or support content, sometimes under changed terms and defaults that security teams never reviewed. This creates a hidden governance gap because the organisation believes the platform is approved, while the AI function inside it is not separately assessed. The result is a control split between the original software approval and the later AI capability that now handles sensitive information.

Practical implication: Re-review SaaS approvals whenever AI features, copilots, or data-processing terms change.

Threat narrative

Attacker objective: The attacker, or the shadow AI operator, seeks hidden access to enterprise data, workflows, and privileged actions without enterprise governance or detection.

  1. Entry occurs through personal AI accounts, unapproved browser extensions, SaaS copilots, or hardcoded API keys that bypass enterprise approval and monitoring.
  2. Credential or permission abuse follows when OAuth grants, saved credentials, or broad extension permissions allow the AI to access mail, files, calendars, code, or internal systems.
  3. Impact emerges through silent data exposure, unauthorized API use, policy violations, and decision errors that are difficult to trace back to the AI activity.

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 an inventory failure before it is a policy failure. Security teams cannot govern what they cannot see, and the article’s examples show that AI use often sits inside personal accounts, browser sessions, or embedded SaaS features. That makes discovery the foundational control, not a supporting one. The practical conclusion is that governance programmes must begin with continuous AI inventory and ownership mapping.

Autonomous AI creates a new access category that conventional IAM does not yet model well. When an agent acts through a human account, the organisation gets attribution without real operational clarity. That is a governance blind spot, because approval workflows, recertification, and offboarding assume a person is the actor. The right lens is to treat active agents as governed entities with explicit scope, lifecycle, and audit boundaries.

Shadow AI governance debt is now a board-level exposure, not a niche technology issue. The article’s cited breach and adoption data point to a pattern where adoption outpaces policy and monitoring. Once AI features are embedded into approved platforms, enterprises can accidentally create high-trust data paths without revisiting the control design. Practitioners should expect governance questions to move from tool approval into lifecycle control, third-party review, and accountability.

The enterprise boundary has shifted from the endpoint to the AI-mediated action. Traditional controls focus on devices, logs, and SaaS approvals, but shadow AI can use legitimate tools to transform authorised access into unauthorised use. That means the real control objective is to govern the AI action path, not just the application catalog. Teams should align their control model to identity, permissions, and observability across the full AI workflow.

Shadow AI is now an NHI governance issue wherever agents use credentials or tokens. Once an AI system can query databases, send mail, or write to shared drives through OAuth or API keys, it behaves like a non-human identity that needs ownership, scope, and offboarding discipline. That is where NHI governance and agentic AI security meet. Practitioners should manage these agents as identities, not as informal productivity helpers.

From our research library:

What this signals

Shadow AI discovery is becoming a governance control, not a reporting exercise. The main failure mode is not that organisations lack AI policy language. It is that they lack a dependable inventory of where AI is embedded, who owns it, and what data it can touch. Inventory-to-control gap: when discovery is incomplete, approval, review, and offboarding all become partial controls instead of enforceable ones. According to the 2026 Infrastructure Identity Survey, 67% of organisations still rely heavily on static credentials despite the risks they pose to agentic AI deployments.

Agentic AI changes the identity model because actions are no longer tied cleanly to a human operator. Once an agent can act through OAuth grants, API keys, or embedded permissions, the enterprise needs explicit ownership and lifecycle controls for the machine actor itself. That makes AI governance a subset of identity governance, not a separate side programme. According to the 2026 Infrastructure Identity Survey, only 13% of organisations feel extremely prepared for the reality of agentic AI despite the majority racing toward autonomous adoption.

For practitioners

  • Build a complete AI tool and agent inventory Map approved, tolerated, and prohibited AI tools across browsers, SaaS apps, extensions, local models, and agents, then assign an owner and data classification to each entry.
  • Revoke unmanaged personal-account access paths Block or step up controls for personal AI accounts, unmanaged browser extensions, and unsanctioned OAuth grants that can move enterprise data outside policy.
  • Separate AI permissions from ordinary SaaS approval Reassess SaaS products whenever copilot features, embedded assistants, or data-processing terms change so that AI functions receive their own review.
  • Treat agents as governed non-human identities Record each agent’s owner, allowed systems, credential source, and termination trigger so offboarding and access review work at lifecycle level.
  • Log AI-mediated actions separately from user activity Preserve prompts, tool invocations, and downstream system actions in a way that distinguishes automated behaviour from ordinary user events.

Key takeaways

  • Shadow AI becomes dangerous when normal user workflows hide AI tools, agents, and data flows from governance.
  • The article’s evidence shows that adoption is already outpacing policy, inventory, and monitoring across many enterprises.
  • The most effective control shift is from reactive discovery to lifecycle ownership for every AI tool and agent.

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.
  • 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.
  • 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.
  • Non-Human Identity (NHI): A digital identity assigned to a non-human entity such as a software application, service account, API key, bot, machine, or AI agent that enables it to authenticate and interact with systems without direct human involvement. NHIs now outnumber human identities in most enterprises by 25 to 50 times.

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. It helps practitioners align identity controls with the broader security programme that now has to govern AI agents too.
NHIMG Editorial Note
Published by the NHIMG editorial team on August 2, 2026.
Updated on October 12, 2026.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM, and Agentic AI security. nhimg.org