AI governance shadow AI SOC 2

Shadow AI: Find and Govern Unapproved AI Tools

Shadow AI explained for SaaS teams: where unapproved AI tools hide, how to discover them, and a practical way to govern them without banning AI.

By Mukesh Bhakar · · 3 min read

Shadow AI is any AI tool your staff use without security or IT review — browser extensions, meeting bots, personal chatbot accounts, AI features switched on inside SaaS apps. It's usually well-intentioned, but it sends company and customer data to places you haven't assessed. The fix isn't a ban; it's discovery, a fast approval path and a short list of approved tools people actually want to use.

Why shadow AI is a real risk

  • Data leakage — customer data, source code, contracts and credentials pasted into tools with unknown retention or training terms
  • Contract breaches — customer agreements often restrict where their data can be processed
  • Audit gaps — SOC 2 auditors and enterprise buyers increasingly ask how you control AI tool usage
  • Inaccurate output — unreviewed AI content reaching customers or decisions
  • Over-privileged integrations — AI apps granted broad OAuth access to email, drives or code

Where shadow AI hides

Place Examples
Personal accounts Staff using personal chatbot logins for work
Browser extensions Writing assistants, summarisers, "chat with this page" tools
Meeting tools AI note-takers joining calls automatically
SaaS features AI assistants enabled inside CRM, docs, ticketing or design tools
OAuth-connected apps AI tools granted access to email, calendars, drives or repositories
Developer tools AI coding assistants, CLI tools, API keys on personal cards

How to discover it

  1. Review OAuth grants in your identity provider and workspace admin consoles — look for AI apps with access to mail, files or code.
  2. Check browser extension inventories through your device management tooling.
  3. Review SaaS admin settings for AI features that are on by default.
  4. Check expenses and card statements for AI subscriptions.
  5. Look at DNS or web proxy logs for traffic to AI services, where you have that visibility.
  6. Ask people. A short, blame-free survey ("which AI tools help you work?") often finds more than technical discovery, and tells you which tools to approve.

How to govern it without killing productivity

Approve good tools quickly

Pick a small set of approved AI tools with company accounts and enterprise data terms — no training on your data, limited retention, SSO. If people have a good approved option, most shadow use disappears.

Make the request path fast

A simple form and a decision within days. Review data use, retention, training terms, security assurance and the access the tool needs. A slow process creates more shadow AI.

Set clear data rules

Spell out which data classes can go into which tools — and that credentials, keys and restricted personal data never go into any AI tool. Our AI acceptable use policy guide has a ready-made table.

Put technical guardrails where they matter

  • Restrict OAuth consent so users can't grant AI apps broad access without approval
  • Manage browser extensions on company devices
  • Turn off AI features in SaaS apps until reviewed
  • Apply data loss prevention to the highest-risk data where you can

Train and keep reviewing

Include AI in onboarding and annual awareness training, and review the approved list quarterly — AI tools change fast.

How this maps to SOC 2

Shadow AI control SOC 2 area
Approved AI tool list and request process Vendor management (CC9)
OAuth and extension restrictions Logical access (CC6)
AI acceptable use policy and training Control environment, communication (CC1, CC2)
Discovery reviews Monitoring (CC4)
AI data incidents in your playbook Incident response (CC7)

See AI governance for SaaS startups for the wider program and AI vendor risk assessment questions for reviewing tools.

FAQ

Should we block public AI tools?

Restrict them for confidential data and give people an approved alternative. Blocking without an alternative pushes usage onto personal devices where you have no visibility.

How often should we look for shadow AI?

Run discovery at least quarterly, and check OAuth grants and new SaaS AI features monthly if you can.

Is a policy enough?

No. A policy sets expectations, but you need discovery, an approved tool list and a few technical controls to make it real — and auditors will look for evidence of all three.

Need help getting AI tool usage under control? Book a free scoping call or see our AI governance service.

Ready to get audit-ready — and AI-ready?

Book a free 30-minute scoping call. You'll leave with a clear view of your gaps, timeline and realistic budget — no obligation.