Shadow AI

Shadow AI is employees using AI tools, such as chatbots, writing assistants or browser extensions, for work without the business's approval, often by pasting in customer data, code or confidential documents.

BeginnerCategory: AI and emerging threats

Also called: unsanctioned AI, unauthorized AI use, BYOAI, bring your own AI

What is shadow AI?

Shadow AI is the AI version of shadow IT. Employees discover that AI tools save them hours, so they use them, often with free personal accounts and without telling anyone. They paste in customer emails to draft replies, upload spreadsheets for analysis, share code for debugging, or install AI browser extensions and meeting note-takers that can read everything on screen or in a call.

Usually nobody means harm. The problem is that company data ends up in services the business hasn't vetted.

Why it matters

IBM's 2025 Cost of a Data Breach study found that one in five organizations reported a breach tied to shadow AI, that organizations with high levels of shadow AI saw breach costs about $670,000 higher on average, and that only 37% had policies to manage AI or detect shadow AI.

The main risks:

  • Data exposure: consumer AI services may keep prompts and, depending on settings, use them to improve their models. Confidential information can also leak through the AI application itself. See sensitive information disclosure.
  • Compliance violations: putting health, payment or personal data into unapproved tools can breach HIPAA, PCI DSS, privacy laws or client contracts.
  • Over-permissioned add-ons: AI extensions and integrations may get access to email, files and calendars.
  • Wrong answers in real work: unreviewed AI output in contracts, code or customer advice.
  • No visibility: you can't protect or respond to what you don't know about.

How to manage it

Banning AI rarely works; people use it on their phones instead. A better approach:

  1. Provide approved AI tools with business accounts, where data isn't used for training and admins have control.
  2. Write a clear AI policy, or add it to your acceptable use policy: which tools are approved and what data must never go in.
  3. Discover what's in use through sign-in logs, OAuth app reviews, browser extension inventories and network logs.
  4. Restrict risky integrations by requiring admin approval for apps connecting to Microsoft 365 or Google Workspace.
  5. Use [data loss prevention](/glossary/data-loss-prevention/) controls to flag sensitive data going to AI sites.
  6. Train staff on safe AI use.

See AI security governance for the bigger picture.

Related terms

Sources

  1. Cost of a Data Breach Report 2025 (IBM)
  2. LLM02:2025 Sensitive Information Disclosure (OWASP Gen AI Security Project)

Drafted with AI assistance, then edited and fact-checked by a person. How we use AI.