Shadow AI and Data Leakage
Your people already use AI. The open questions are whose account it's on and where the data goes.
The lines under "How it sounds" are typical phrasing we wrote, not quotes from clients or interviews.
How it sounds
- “I paste the client file into my ChatGPT and it writes the summary.”
- “I built a little tool over the weekend that pulls from our customer list.”
- “Which plan? The free one, I think.”
What it looks like
Staff use personal AI accounts for work and paste in client files, contracts, pricing and code. Some build their own apps with AI coding tools, connected to company data. Nobody knows which of those personal plans can use what's typed into them to train the provider's models.
Why it happens
People have real work, and the tools are a browser tab away. When the company hasn't provided approved tools or written rules, people fill the gap themselves, and most of them are trying to do good work faster.
What it costs
- Scale. Microsoft and LinkedIn's 2024 Work Trend Index found "78% of AI users are bringing their own AI tools to work," and 80% at small and medium-sized companies.
- Breaches. IBM's 2025 Cost of a Data Breach study found "security incidents involving shadow AI accounted for 20% of breaches," organizations with high levels of shadow AI saw "an average of $670,000 in higher breach costs," and "63% of breached organizations either don't have an AI governance policy or are still developing a policy." The sample is 600 organizations that had a breach, and IBM sells security.
- Model training. OpenAI's help center says that for services for individuals, like ChatGPT, "we may use your content to train our models," with an opt-out, while business plans and the API aren't used for training by default. Anthropic's consumer plans use chats for training if the user allows it. Google tells Gemini app users, "Please don't enter confidential information that you wouldn't want a reviewer to see or Google to use to improve our services." The safe public wording is "can be used to train models unless you turn it off" or "if you allow it."
First move
Move it into the open instead of banning it. Ask what people use and why, put work use on business plans the company controls, and write one page on what data stays out. Then pick the best home-built tools and make them supported ones.
The fifteen patterns, with what each one costs and a first move, are in The AI anti-pattern field guide.
Related AI patterns
- AI Left to Emerge. No decisions made on purpose, so whatever grows in the dark is what you get.
- The Quiet Users. People use AI every day and don't say so.
- The Unchecked Draft. It looks finished, and someone downstream has to make it right.
An AI strategy decides who owns the work, which tools stay, and what changes first.