AI discovery is the missing first step in enterprise governance

At a glance

What this is: Holistic AI argues that AI discovery is the prerequisite control for governing hidden, duplicated, and unmanaged AI across the enterprise.

Why it matters: For IAM, security, and governance teams, incomplete AI discovery leaves ownership, accountability, and risk classification built on assumptions rather than an enforceable inventory.

Context

AI governance fails first at visibility. If teams cannot see which models, copilots, chatbots, plugins, RAG systems, or embedded AI features are running across the business, every downstream control starts from an incomplete inventory. In identity terms, that means ownership, access boundaries, and lifecycle decisions are being made against unknown assets rather than governed systems.

This article is about the governance gap between what leadership thinks exists and what the environment actually contains. Holistic AI frames discovery as the first operational control because the absence of discovery leaves shadow AI outside policy, monitoring, and accountability. That pattern is typical in fast-moving enterprises, especially where AI is embedded in SaaS, developer tooling, and data platforms.

Key questions

Q: What breaks when AI agent discovery is incomplete?

A: When discovery is incomplete, the organisation cannot know which agents exist, what they are connected to, or what they can access. That leaves policy enforcement, enrichment, and review operating on a partial inventory, which is the same as governing only part of the environment. Hidden agents become hidden access paths.

Q: Why does shadow AI create risk even when employees are trying to be productive?

A: Because the risk is not intent, it is uncontrolled data movement. A user who pastes sensitive material into an unsanctioned AI service can expose information outside enterprise policy, retention, and compliance boundaries even if the work request was legitimate.

Q: How should security teams govern AI features embedded in SaaS applications?

A: Treat embedded AI as a machine identity problem with data access implications. Inventory the feature, map the connected permissions, define what data it may use, and monitor retention and sharing paths. If the AI feature can read corporate content, it needs explicit approval, logging, and periodic review like any other privileged integration.

Q: How do organisations know whether AI governance is actually working?

A: AI governance is working when teams can prove that data access, identity permissions, and runtime controls line up with policy in practice. A useful test is whether the organisation can answer who accessed what, through which identity, and whether any out-of-policy movement was blocked or detected in time.

Technical breakdown

Why AI discovery has to happen before inventory

Discovery is the automated identification of AI artefacts across environments, while inventory is the managed record built from that evidence. Without discovery, inventory is just a spreadsheet of self-reported systems, and self-reporting misses embedded copilots, ad hoc chatbots, and hidden workflows. The article’s key point is that governance cannot classify what it has not found, and compliance cannot attest to what is absent from scope. In practice, discovery needs to scan clouds, repos, data platforms, documents, and LLM or agent platforms so the control surface matches reality, not procurement memory.

Practical implication: Treat discovery as the control that establishes scope before any AI governance workflow starts.

How shadow AI emerges inside normal business activity

Shadow AI is not only rogue experimentation. It also appears through ordinary staff behaviour, approved SaaS features, browser extensions, developer assistants, and unreviewed third-party AI capabilities. That makes the problem structurally different from classic asset sprawl because the hidden element may be inside a sanctioned tool rather than a separately procured one. The governance failure is fragmentation: each team sees a local feature, but no one sees the enterprise-wide AI footprint. Identity and access decisions become risky when an AI capability can process data, call APIs, or influence workflows without formal ownership or review.

Practical implication: Map AI features and AI-enabled workflows to owners, data paths, and approval boundaries before they are used at scale.

Why discovery enables ontology, monitoring, and auditability

The article’s Discovery → Inventory → Ontology pipeline shows why discovery is not an isolated scan. Inventory turns found artefacts into metadata-rich assets with ownership and risk context, and ontology maps relationships across datasets, endpoints, prompts, and applications. That relationship layer is what makes policy enforcement and change tracking possible. For governance, this matters because controls must follow behaviour and lineage, not just names on a register. A queryable AI graph creates the basis for continuous oversight, incident triage, and defensible compliance evidence.

Practical implication: Build a governed AI graph so monitoring and compliance can track how each AI asset is connected and who controls it.

Threat narrative

Attacker objective: The practical objective is to exploit unseen AI systems and the data they touch before security and governance teams know the assets exist.

1. Entry occurs when employees, developers, or business teams introduce AI through public chatbots, embedded copilots, extensions, or sand-boxed experiments without central visibility.
2. Credential or data exposure follows when those tools process confidential contracts, source code, customer records, prompts, or internal datasets outside approved governance paths.
3. Escalation happens when hidden AI workflows, RAG systems, or third-party features connect to broader enterprise data and operational processes without logging or review.
4. Impact is governance failure at scale, because risk classification, monitoring, compliance, and accountability are all built on an incomplete inventory.

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

AI discovery is now a governance primitive, not an optional control. The article is right to treat discovery as step zero because every later process depends on an accurate AI asset boundary. If the organisation cannot see its AI footprint, it cannot meaningfully classify risk, assign ownership, or evidence compliance. The practitioner conclusion is simple: inventory quality determines governance quality.

Shadow AI is a policy failure before it is a technical failure. Unreviewed copilots, extensions, and internal experiments show that AI risk often enters through normal work patterns, not malicious intent. That changes the control question from banning tools to establishing authoritative visibility and lifecycle ownership. The practitioner conclusion is to govern usage paths, not just named products.

AI governance debt is created when discovery lags adoption. Once teams have built workflows, connected data sources, and delegated tasks to AI, the backlog is no longer just remediation. It is a structural deficit in accountability, testing, and auditability that gets more expensive over time. The practitioner conclusion is to reduce unknowns early rather than trying to certify a moving target.

Discovery must be continuous because AI behaviour changes faster than policy cycles. Static inventories miss new plugins, embedded features, prompt logic, and agent workflows as soon as teams change tools or data paths. That is why discovery has to feed an ongoing control model rather than a one-time project. The practitioner conclusion is to make visibility operational, repeatable, and tied to change management.

Identity and governance intersect whenever AI can act on data or systems. The moment an AI feature can call APIs, access documents, or trigger workflows, identity, privilege, and ownership questions appear. That is where NHIMG’s lens matters: the organisation is not only managing AI, it is managing who or what is authorised to use AI capabilities. The practitioner conclusion is to align AI discovery with identity governance from the start.

From our research library:

• 52% of respondents see AI security decision-making power shifting toward platform and infrastructure teams rather than the executive suite, according to the 2026 Infrastructure Identity Survey.
• 63% of organisations surveyed lacked AI governance policies to manage AI or prevent shadow AI, according to IBM's 2025 Cost of a Data Breach Report.
• Read next: Shadow AI and AI Agent Discovery Guide

What this signals

AI governance debt: when discovery trails adoption, organisations accumulate unowned copilots, hidden workflows, and unreviewed third-party features that become harder to classify or retire over time. The remediation problem is not just visibility, it is the backlog of decisions that visibility was supposed to prevent.

Discovery has to be treated as an identity and access control issue as much as a governance issue. Once AI can touch documents, APIs, or operational workflows, the question becomes who or what is authorised to use those capabilities, and whether that authority still matches the current business need.

For practitioners

• Automate AI discovery across all business surfaces Scan clouds, repositories, data platforms, documents, endpoints, and AI-enabled SaaS to identify every live AI artefact before governance decisions are made.
• Assign ownership to every discovered AI asset Require a named business owner, technical owner, and approval path for each model, copilot, workflow, or embedded feature found in discovery.
• Classify AI by data access and business criticality Tag each discovered AI asset with the data it can reach, the workflow it affects, and the risk tier it belongs to so approvals become repeatable.
• Connect discovery outputs to change management Reconcile discovery results against procurement, SDLC, and SaaS change records so new AI capabilities are not invisible to control owners.
• Use discovery evidence to drive policy enforcement Block unapproved AI assets, monitor approved ones continuously, and require revalidation whenever prompts, data sources, or integrations change.

Key takeaways

• AI governance fails when discovery is treated as a later-stage compliance activity rather than the control that establishes scope.
• Hidden copilots, extensions, and sandboxed experiments can create shadow AI even inside approved platforms, which makes inventory quality a security issue.
• The practical fix is continuous discovery tied to ownership, classification, and change management so AI controls can follow the actual environment.

Key terms

• AI Discovery: AI discovery is the process of automatically finding AI tools, embedded features, agents, and integrations operating in an environment. It provides the first visibility layer for governance, but it does not by itself explain ownership, permissions, or risk.
• 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.
• API Governance Debt: API governance debt is the accumulation of weak, inconsistent, or missing controls around how APIs are designed, approved, documented, secured, and retired. It creates hidden risk over time. Technically, it includes gaps in authentication, authorization, versioning, inventory, ownership, logging, and policy enforcement that make APIs harder to trust and govern.
• AI Ontology: A structured map of relationships between AI assets such as datasets, prompts, pipelines, applications, and deployment endpoints. It helps teams understand dependencies, ownership, and downstream impact, which turns a static inventory into a governed and queryable view of the AI estate.

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 designed for practitioners who need to connect identity governance to emerging machine and agent risk.