Shadow AI: What It Is, Why It's Risky, and How to Manage It
Updated · EraseAI
Shadow AI is the use of AI tools at work without the knowledge or approval of IT and security teams: a personal ChatGPT account used for client emails, a free AI transcription app in sales calls, an AI browser extension a developer installed last week.
It is the AI-era version of shadow IT, and it is common because the tools are free, useful and one click away. IBM's Cost of a Data Breach Report 2025 found that one in five organizations it studied had a breach involving shadow AI, and that those breaches cost more than average.
Why people use unapproved AI
- The approved tool doesn't exist yet, or is slower to get than a free sign-up.
- The personal account is already open in another tab.
- People don't know which data is sensitive, or assume the AI provider deletes it.
- A ban was announced, but the work still needs doing.
What can go wrong
- Data leaves through personal accounts with consumer terms, where conversations may be used for training and are outside your retention, legal hold and deletion processes.
- No record exists of what was shared, which makes breach assessment and regulatory notification guesswork.
- Compliance obligations are broken quietly, for example sending personal data to a processor without a data processing agreement under GDPR.
- Unvetted apps get broad access through browser extensions and OAuth connections to email and drives.
A five-step plan
- Discover. Survey teams and review browser extensions, OAuth grants and web traffic to AI domains. Expect to find more than you thought.
- Approve good options fast. Give people a sanctioned AI tool on a business plan. Shadow AI shrinks when the approved path is easier.
- Set clear rules. Publish a one-page acceptable use policy: approved tools, forbidden data, who to ask.
- Check at the point of use. A browser and mobile AI firewall catches secrets and personal data in any AI tool, approved or not, at the moment of sending.
- Measure and adjust. Track what is caught and where, then fix the process behind repeated findings.
Why bans don't work
Several large companies restricted ChatGPT in 2023 after staff pasted confidential material into it; Samsung's restriction after engineers shared source code is the best-known case. Bans buy time, but usage tends to move to personal phones, where you can't see or protect it. Approving tools and checking what is sent to them is more durable.
Check every message before it reaches AI
EraseAI stops API keys, passwords, card numbers and personal data in ChatGPT, Claude and Gemini. Free in Chrome, no account needed.
Frequently asked questions
Is shadow AI the same as shadow IT?
It is a subset. Shadow IT covers any unapproved technology; shadow AI is specifically unapproved AI tools and features, which carry extra risk because people paste data into them freely.
How do I find shadow AI in my organization?
Combine an anonymous survey with technical discovery: browser extension inventories, OAuth app grants in your identity provider, and web proxy or DNS logs for AI domains.