It’s month-end, and you’re working through a client’s card feed. Between the fuel and the office supplies, there’s a ChatGPT subscription. A few lines down, an AI note-taker you’ve never heard of. Then, buried in the expense claims, a writing assistant someone bought on their personal card and claimed back.
None of these went through anyone. The owner probably doesn’t know about half of them. And if you asked the business which AI tools it uses, you’d get a shrug, or a list that’s missing most of what’s in front of you.
A decade ago, we called this shadow IT: software that crept into a business without anyone signing off on it. It’s back with a new name and much higher stakes. And this time, accountants and bookkeepers are among the first people in a position to see it.
Shadow AI: the same problem, with higher stakes
Shadow AI is any AI tool used for work without the business knowing about it or signing off on it. That covers paid subscriptions on a company card, free tools someone signed up for with their work email, and AI features quietly switched on inside software the business already pays for.
On the surface, it looks like the old shadow IT problem. The difference is what these tools do with what you give them. A rogue project-management app holds task lists. An AI tool reads whatever gets pasted into it, whether that’s a customer list, a payroll export, a supplier contract or last quarter’s management accounts. It may store that data, or learn from it, on servers nobody at the business has ever vetted.
And where shadow IT crept in over years, shadow AI has arrived almost overnight.
Half the workforce is already doing it
Pick almost any recent survey and the numbers land in the same place. BlackFog’s research of 2,000 UK and US employees, published in January, found that 49% use AI tools their employer hasn’t approved. More than half had connected AI tools to other work systems without IT ever knowing. In finance teams specifically, TrustedTech found that 59% of employees use unapproved AI, even though 80% said they understood the risks.
It isn’t just junior staff experimenting, either. TrustedTech’s research also found that 65% of decision-makers use shadow AI, roughly double the rate of the people who report to them. When the owner is the one signing up for new tools, there’s nobody above them to notice.
One caveat is worth saying out loud. Most of this research looks at larger companies, the kind with IT and security teams. Your small and mid-sized clients usually have neither, so there’s little reason to think the picture is any tidier inside a 20-person business.
Why it lands on your desk first
An IT team can go months without spotting a browser-based AI tool. You can’t, because somebody has to pay for it.
AI tools are designed to be bought without asking. Plans are priced per person, per month, usually low enough that nobody needs sign-off. Free tiers become paid ones the day a user hits a limit. Staff pay on personal cards and claim it back, so the vendor never appears as a supplier. Software the business already uses adds AI tiers and extra seats, folded into an invoice that gets approved without a second look. And usage-based tools bill a different amount every month, so they don’t even look like subscriptions.
Each of those lines is small. Added up, they’re a software stack nobody chose, sitting inside your client’s general ledger. And you’re often the only person who sees all of it in one place.
What’s actually at risk
It’s tempting to treat this as a tidy-up job: a handful of forgotten subscriptions to cancel. The cost turns out to be the smallest part of the problem.
Start with the money anyway, because it’s what gets a client’s attention. Shadow AI rarely means one big bill. It means three different AI note-takers across three teams, seats nobody has opened in months, and annual plans that renewed automatically after the person who bought them left. No single line is worth chasing, which is exactly why nobody owns the total.
The bigger exposure is data. IBM’s 2025 Cost of a Data Breach Report found that one in five organisations had suffered a breach linked to shadow AI. Where shadow AI use was widespread, breaches cost an average of $670,000 more, and nearly two-thirds of those incidents exposed personal information. For a client in professional services, the data at risk might not even be theirs. It might belong to their own clients.
Then there’s compliance. A business that doesn’t know which AI vendors it uses can’t say where its data lives, or whether it’s being used to train someone else’s model. That becomes a real problem the moment a client fills in a cyber insurance renewal, answers a customer’s security questionnaire, or needs to account for the processors handling personal data under GDPR. It’s why Gartner expects more than 40% of organisations to suffer a security or compliance incident caused by unauthorised AI by 2030.
Where to look next time you’re in the ledger
You don’t need a security audit to get a first picture. Next time you’re in a client’s books, give it twenty minutes:
- Search transactions for the obvious names: OpenAI, ChatGPT, Anthropic, Claude, Perplexity, Midjourney, Jasper, Otter, Fireflies and Grammarly.
- Group card spend by payee and look for small, identical charges that repeat every month.
- Check expense claims coded to software or subscriptions, which is where personally bought tools tend to hide.
- Read the line items on existing software invoices for anything labelled AI, Copilot or Assistant that wasn’t there last year.
- Watch for the same vendor across several cards, a sure sign that people are buying their own seats.
Then do one thing that makes every future month easier: give AI tools their own account or tracking category under software. Once AI spend has its own line, it can’t hide anymore.
Don’t ban it. Bring it into the open.
When clients see the list, the instinct is often to shut everything down. That rarely works. In TrustedTech’s research, nearly a third of employees said they’d keep using AI tools even if their employer banned them. A ban simply pushes the usage onto personal accounts, where it’s even harder to see.
A short, practical conversation works better. Walk the owner through what you found: the tools, who’s paying for them, and what it all adds up to over a year. Ask what data goes into each one, because that question often matters more than the price. Where several teams have picked different tools for the same job, help them settle on one and move to a business plan with proper admin controls.
Then agree a one-page policy. It only needs to cover who can buy AI tools, what should never be pasted into them, and where to find the approved list. Keep it short enough that people actually read it.
Finally, come back to it every quarter. AI tools appear, get replaced and renew faster than any software category before them. A regular check catches the new ones before they quietly become next year’s renewal.
Your next advisory conversation is already in the books
Most clients will never ask you about AI risk. But almost all of them will listen when you show them money they didn’t know they were spending, and data going places they didn’t know it went.
That’s what makes shadow AI such a natural opening. You already have the transaction data, and you already have the trust. A regular review of software and AI spend turns a one-off discovery into an ongoing service. It also positions your practice as the adviser who understands how your clients’ businesses actually run today.
AppVentory is built to help accountants and bookkeepers do exactly that across their whole client base, without trawling through ledgers by hand. It brings each client’s software and AI subscriptions into one view, shows which tools are actually being used, and highlights what to keep, consolidate or cancel.
Frequently asked questions
What’s the difference between shadow IT and shadow AI?
Shadow IT is any software used without approval. Shadow AI is the part of it that processes whatever people type or upload. That raises the stakes, because sensitive information can leave the business in a single prompt.
Is shadow AI always a bad thing?
Not at all. It usually means staff have found tools that genuinely make them faster. The risk comes from nobody knowing the tools exist, so nobody checks the cost, the data terms, or whether a better-controlled option is available.
How do I find AI subscriptions in a client’s accounting software?
Search transactions for common AI vendor names, review recurring card charges and expense claims coded to software, and check existing software invoices for new AI add-ons. Then set up a dedicated account or tracking category so AI spend stays visible from then on.
Should businesses ban unapproved AI tools?
Bans tend to push usage onto personal accounts, where it’s even harder to see. A short policy, an approved list and a regular review work far better than a blanket ban.
Conclusion
Shadow AI isn’t going away, and it isn’t waiting for anyone’s permission. The businesses that handle it well won’t be the ones that ban it. They’ll be the ones that can see it. As the person who already sees every transaction, you’re in the best position to help them get there. Start with the next ledger you open.
Book a demo with us today.
