How can organisations reduce exposure from unmanaged SaaS and AI tools?

Why This Matters for Security Teams

Unmanaged SaaS and AI tools create exposure because they often sit outside formal procurement, identity, and logging paths. That means sensitive data can move into applications that never passed security review, never inherited SSO enforcement, and never enter the normal offboarding process. Current guidance suggests prioritising the tools most likely to handle regulated data, credentials, or customer content, then treating discovery as the start of control, not the finish.

NHIMG research on the Guide to the Secret Sprawl Challenge shows how fragmented secret handling and weak lifecycle control quickly turn ordinary tool adoption into persistent exposure. That pattern is consistent with broader industry reporting in the NIST Cybersecurity Framework 2.0, which reinforces that asset visibility, access governance, and continuous monitoring have to work together.

The practical problem is not only that shadow tools exist, but that users keep connecting them to email, file stores, code repositories, and AI assistants long after the business need has changed. In practice, many security teams encounter data leakage only after a shared link, exposed token, or unsanctioned AI plugin has already widened the blast radius, rather than through intentional review.

How It Works in Practice

Reducing exposure starts with discovering which SaaS and AI tools are active, which identities can reach them, and which data they touch. A useful triage model is to rank tools by business criticality, authentication path, data sensitivity, and whether they can bypass central controls such as SSO, SCIM, or PAM. For AI tools, also include prompt retention, connector permissions, and whether the system can ingest files, source code, or customer records.

Once the inventory is reliable, organisations should tie it to lifecycle actions: de-provision unused accounts, revoke API keys and OAuth grants, and require periodic review of high-risk tools. For unmanaged SaaS, that often means integrating discovery with CASB, identity governance, and cloud access monitoring. For AI tools, the same discipline applies to browser extensions, embedded copilots, and standalone chat systems that may quietly store or reuse data. NHIMG’s 52 NHI breaches Report and Ultimate Guide to NHIs — Lifecycle Processes for Managing NHIs both reinforce that lifecycle control matters as much as detection.

A practical control pattern looks like this:

• Discover SaaS and AI usage from identity, endpoint, browser, and network telemetry.
• Classify tools by sensitivity, authentication strength, and data exposure risk.
• Remove or quarantine tools that cannot support SSO, logging, or admin review.
• Automate offboarding for accounts, tokens, integrations, and shared workspaces.
• Review high-risk tools on a fixed cadence and after role changes or exits.

Where this becomes effective is when security owns the removal path, not just the inventory view. These controls tend to break down when business units can self-provision apps through personal email and external AI services because the organisation loses both authority over credentials and reliable evidence of data movement.

Common Variations and Edge Cases

Tighter control over SaaS and AI access often increases friction, requiring organisations to balance user convenience against reduction in unreviewed exposure. That tradeoff is especially visible in product, engineering, and marketing teams that adopt tools quickly to move fast. Best practice is evolving, but there is no universal standard for how aggressively to block unsanctioned AI usage while still supporting innovation.

Some environments should focus first on containment rather than full elimination. For example, if a business depends on a narrow set of external collaboration tools, the safer move may be to enforce SSO, restrict file uploads, and monitor data export rather than trying to ban the category outright. The same is true for AI assistants used in development workflows: the immediate goal is often to control connectors, tokens, and prompt retention before attempting broader policy changes. The Top 10 NHI Issues is useful here because it highlights how secret sprawl, over-permissioning, and weak governance compound across tool categories.

One important edge case is third-party AI embedded inside approved SaaS. Even when the primary application is sanctioned, downstream AI features can introduce new retention or training risks that were not covered in the original review. Another is acquired companies, where duplicate SaaS stacks and orphaned identities are common. In both cases, discovery without remediation simply documents the problem. In high-churn environments, unmanaged tools multiply faster than review cycles unless access removal is automated.

Related resources from NHI Mgmt Group

• How can organisations reduce hidden risk in shadow SaaS and unmanaged entitlements?
• How can organisations reduce data exposure in AI tools?
• How should organisations reduce risk from unmanaged access privileges?
• What breaks when organisations rely only on blocking unapproved AI tools?