Shadow AI is AI being used for work that the business never approved. That’s the whole definition — not a product, not a type of attack, just the gap between what your team uses and what you’ve agreed to. It matters because you can’t write rules for tools you can’t name, and in most small businesses nobody has been asked to name them.
It is also, almost always, motivation rather than mischief. Somebody had a task to finish, the approved route was slower or didn’t exist, so they found a faster way. Nobody is hiding anything. Nobody ever asked.
Why it shows up without anyone deciding to let it
Two things make this different from the usual unapproved-software problem.
The first is how little friction there is. Signing up takes a minute and a work email address — nothing installed, no ticket raised, no card charged on a free tier. The usual trip-wires that tell you new software arrived never fire.
The second is that the useful thing and the risky thing are the same action. Pasting a client contract in for a quick summary is someone trying to work faster, and it is also the moment company information leaves your control. They weren’t careless; they were efficient, with no rule telling them where the line was.
The four places Shadow AI hides
Most owners can name one AI tool. Ask the team, and the list gets longer.
1. Tools people signed up for themselves
The obvious one, and usually the smallest. Somebody created an account on a chat assistant, a writing tool or a transcription service. A personal account on a consumer plan generally means no administrative visibility for the business, no control over its settings, and no way to recover it if that person leaves.
One thing worth being precise about: what any tool does with what you type into it depends on the tool, the plan and the settings. Consumer and business or enterprise tiers are usually configured differently, and the specifics change — so check the vendor’s current terms for the tool you’re considering, rather than a summary you read somewhere, including this one.
2. AI features inside software you already pay for
The big one, and the one people forget, because nobody installed anything. Your office suite, CRM, help desk and note-taking app have all been adding AI summarising, drafting and search features. Some arrive switched on. Some appear in a side panel one Tuesday morning.
Nobody is being furtive — they’re using a feature of software you bought. But “what information is allowed to go into an AI tool?” applies to a summarise button in software you own exactly as much as to a website somebody signed up for.
3. Meeting recorders
Transcription bots that join a call and take notes are now ordinary, and they create two problems.
The first is what the recording holds: a half-hour meeting can contain client names, commercial terms, staff matters and figures nobody would have put in writing. The second is consent, and it isn’t uniform — recording-consent law varies by state, California is generally treated as an all-party-consent state, and Tennessee generally as one-party. That’s worth confirming with your own counsel rather than with a blog post. Verify with counsel; this is not legal advice.
And the bot isn’t always yours — it can arrive with a participant from another company, which makes “whose recorder is this, and who agreed to it?” a question your team needs answered before the call starts.
4. Browser add-ons
Extensions that summarise or rewrite whatever page is open have to be able to read whatever page is open. That’s not a flaw — it’s how they work. But the extension that summarises an article is also sitting there while somebody reads a mailbox, a payroll screen or a client portal.
Extensions install in a few clicks on a personal browser profile, so they rarely appear in any inventory — added once for one task, still there a year later.
What to do about it, in order
Three steps, and only the last is technical.
One: find out what’s in use. Ask your team which AI tools they already use for work, and make clear there’s no penalty for answering honestly. That list is the starting point for everything that follows — rules written without it are guesses. There’s a method for this in Shadow AI at work, including what an administrator can review in your tenant.
Two: make the decisions. Which tools are approved, and for what. What information can go into them, and what has to come out first. Who answers “can I use this?”, and how fast. Who checks AI-assisted work before a client sees it. Who to tell when something goes wrong. Business decisions about how much risk you’ll carry, not IT settings.
Three: put the decisions into your systems. Approved tools as company accounts with sensible settings and access limited to each role — see AI governance and AI use policy services. If anyone has connected an AI tool to a company mailbox, start with should you connect AI to your email?; that one click grants more than people expect.
Written rules are cheap air cover for your staff. Nobody should be in trouble for asking first.
Three questions for your next team meeting
- What AI tools are we using, and for which tasks?
- What information is allowed to go into them — and what has to come out first?
- Who answers “can I use this?”, and how fast?
If those answers come back fuzzy, that isn’t a people problem — it’s a page of rules nobody has written yet. Once they exist, they belong alongside the rest of your security awareness training, not in a folder nobody opens.
Frequently Asked Questions
Is Shadow AI the same thing as Shadow IT?
Shadow AI is a subset of Shadow IT that behaves differently in two ways. It needs no installation, so the signals that new software arrived don’t fire — and it often arrives as a feature of software you already approved, so there’s nothing to “catch”. Blocking installs and reviewing expense claims misses most of it.
Should we just ban AI tools at work?
You can, and in some regulated work that’s reasonable, but it’s harder to hold than it looks. A ban tends to move the usage onto personal accounts and devices where you can’t see it, leaving the same exposure with less visibility. If you do want a prohibition, write it down, explain why, and give people a route to request exceptions — an unwritten ban is just an assumption.
How do I find out what AI tools my employees are using?
Ask them first, with an explicit no-penalty framing — the conversation gets you further than any scan. Then look in the four places above, and have an administrator check which third-party applications have access to your Microsoft 365 or Google Workspace tenant. The fuller method is in Shadow AI at work.
Do small businesses really need an AI policy?
Yes, and it’s shorter than you think — one page naming the approved tools, what information can go into them and who to ask is enough to start. The size of the business doesn’t change the questions, only how long the answers take.
Where do we start if we have no AI rules at all?
Start with the list of what’s in use, make the handful of decisions any workable policy comes down to, then write them on one page and tell your team what changed. If you’d rather talk it through, we run a free 30-minute Shadow AI Audit call and send an AI Acceptable Use Policy template to adapt afterwards.
Talk it through
We run a free 30-minute Shadow AI Audit call for businesses in Los Angeles, Burbank, Nashville and Franklin — current AI use, the safety points worth passing on to your team, and a policy template to adapt. No obligation, no charge.
Book the 30-minute Shadow AI Audit · or call (818) 284-4118.
Or call (818) 284-4118.
About the author
Garett Chipman - CEO, TVG Consulting
Garett Chipman is an IT executive and entrepreneur with more than 20 years of experience designing business technology solutions. He founded TVG Consulting in 2005 after holding CTO and management roles at companies including GRB Entertainment and AMG, and as CEO he oversees overall client operations, strategic technology decisions, and custom AI and automation design.