Shadow AI is the name that has stuck for AI tools in use at work that the business never approved. It is rarely bad intent. People found tools that save them time, nobody had said which ones were allowed and the free versions took one click. When you find them, the aim is to give the people using them an approved route.
What it usually looks like
- Personal ChatGPT, Claude or Gemini accounts, signed up with a work email address and sometimes paid on a personal card.
- Browser extensions that read the pages people open, including email and internal systems.
- Note-takers connected to calendars that join calls on their own.
- AI features switched on inside software you already pay for, often by default in an update.
- Apps on phones, used for work because they are quicker than a laptop.
Why it appears
Three things together produce it: tools that are free and good, work that is pressured and no written answer to "am I allowed to use this?". Once an approved tool and a short rule remove the last one, most of the rest takes care of itself.
Seven places to look
- Ask. A short question to everyone, about which AI tools they use for work and on which accounts, with a clear statement that nobody is in trouble for the answer. It finds more than any tool, as long as people believe the second half.
- Connected apps. Microsoft 365 and Google Workspace both show which outside apps people have granted access to their work accounts. AI assistants, note-takers and extensions with access to mail, files or calendars appear there.
- Accounts on your domain. Several AI vendors let a business see or claim accounts created with its email domain, which turns a list of personal sign-ups into something you can act on.
- Network and security tools. Cloud app discovery, web filtering and DNS logs show which AI services are being reached from your network and managed devices.
- Managed devices. Device management can list the browser extensions and apps installed on company laptops and phones.
- Expenses and card statements. Subscriptions to AI tools claimed back, or paid on company cards, show where people found something worth paying for.
- Calendars. Note-taker accounts appear as participants in meetings and are easy to spot once you look.
What to do with what you find
Sort each tool into one of three groups:
- Approve it, and move people onto a company account under company sign-in, with its terms checked and its entry on the register.
- Replace it, where an approved tool already does the same job, such as Copilot Chat on a work account in place of a personal chat tool.
- Stop it, where the tool cannot meet your rules, with a short explanation and a pointer to the approved alternative.
Then make the route permanent: a named person to ask, a quick way to request a new tool and a one-page acceptable use policy that says what is allowed. If something sensitive has already gone into a personal account, the incident route applies, and the person who tells you should be thanked for it.
Keep looking
New tools appear every month, so a single sweep goes out of date quickly. Repeat the connected apps check and the question to everyone each quarter, and treat new AI features in your software as new tools, as set out in checking a supplier's use of AI.
Where this fits
Finding what is in use, on whose accounts and reaching what, is exactly what the readiness snapshot does, and it is the first scoped piece of Govern. Watching for new unapproved tools after that is part of Trackside.