
Shadow AI: AI tools used without IT approval
In short: Shadow AI is staff using AI tools and accounts that the organisation has not approved or cannot see, often with company data inside the prompts.
Shadow AI is the use of AI tools, models or accounts inside an organisation without approval or visibility from IT, security or governance. Typical cases: staff pasting company documents into personal chatbot accounts, installing browser AI extensions or connecting unapproved AI apps to work systems. IBM's Cost of a Data Breach Report, released on 30 July 2025, found one in five organisations reported a breach due to shadow AI.
Shadow AI in plain words
People are doing their jobs faster with tools nobody approved, while the company has no idea where its data went.
How it works
It usually starts with a real need and a slow or missing official option. Employees sign up with personal or work emails, use free tiers, install extensions or connect AI apps to email and drives through OAuth. Data flows out through prompts, uploads and connected apps, outside logging, contracts and data-protection terms.
OWASP's LLM02 entry lists what is at stake: personal data, financial details, health records, confidential business data and security credentials. The same IBM report found that organisations with high levels of shadow AI saw an average of $670,000 in higher breach costs, and that only 37% had policies to manage AI or detect shadow AI.
Example
As an illustration, a sales team uses a free AI note-taker that joins every client call and stores transcripts on the vendor's servers. Nobody signed a data-processing agreement, and the client contracts forbid recording without consent.
Where you'll meet it
- Security reviews of browser extensions and OAuth app grants
- Network and endpoint telemetry showing traffic to AI services
- Employee surveys about AI use
- Incident reports after sensitive data appears in an external tool
What it changes for you
For security teams, visibility comes first: discover which AI services are used, by whom, with which data. My view: for leaders, a ban alone rarely works, and approved, easy alternatives with clear rules do more. For employees, knowing which tools are approved for which data protects both them and the company.
Shadow AI vs approved AI
Approved AI runs under contracts, data controls and logging the organisation chose. Shadow AI runs under whatever terms the individual clicked through.
Common mistake
My view: responding with a blanket ban. Usage moves to personal phones and disappears from view entirely. A sanctioned tool that is good enough to use, plus clear data rules, reduces shadow AI far more than a policy email.
Related terms
AI governance · AI gateway · AI-BOM · Prompt injection
Give people an approved tool that is good enough to use, with clear data rules, before you reach for a ban.


