Short answer: Shadow AI is staff using AI tools at work that the firm never approved, most often a personal ChatGPT or Claude account. In a small firm it usually looks like this: someone pastes a client's file into a chatbot to save an hour. If your firm has no approved AI tool, it is very likely already happening. In Microsoft and LinkedIn's 2024 survey, 78% of people who use AI at work brought their own tools, and 80% at small and medium-sized companies. A ban on its own leaves staff with nothing to use instead. What works is a short written policy, an approved tool that is good enough that staff want to use it, and a few rules about what client data may go in.
What counts as shadow AI?
IBM defines it this way: "Shadow AI is the unsanctioned use of any artificial intelligence (AI) tool or application by employees or end users without the formal approval or oversight of the information technology (IT) department" (IBM).
Most small firms have no IT department, so read "IT" as "whoever is responsible for the firm's client data". Usually that is you. In practice, any of these is shadow AI:
- A staff member using a free or personal ChatGPT, Claude or Gemini account for client work.
- An AI note-taker that someone added to client calls on their own.
- A browser extension that rewrites emails or summarizes documents and sends the text to its maker.
- AI features switched on inside software you already pay for, without anyone checking what they keep.
None of this is usually done in bad faith. IBM's own explanation is that "Employees often turn to these tools to enhance productivity and expedite processes." People use them because they work.
How common is it?
More common than most owners think, and much of it is hidden:
- Bring your own AI. "78% of AI users are bringing their own AI tools to work (BYOAI)", and "it's even more common at small and medium-sized companies (80%)". Also: "52% of people who use AI at work are reluctant to admit to using it for their most important tasks" (Microsoft and LinkedIn, 2024 Work Trend Index).
- Sensitive data without permission. A CybSafe and National Cybersecurity Alliance survey of more than 7,000 people found that 38% of employees "share sensitive work information with AI tools without their employer's permission", and 52% of employed participants had not received any training on safe AI use (Infosecurity Magazine, September 2024).
- Company information going in. In Cisco's 2024 privacy study, people said they had entered "employee information (45%) or non-public information about the company (48%)" into generative AI tools (Cisco).
These surveys mostly cover larger employers. We found no survey limited to firms under 25 people.
Why it matters more for a firm that holds client data
A staff member pasting a client's file into a personal AI account creates three problems at once:
- The data has left your control. Personal plans keep chats, and some use them for training by default. On ChatGPT Free, Plus and Pro, "data sharing is enabled by default" until the user turns off "Improve the model for everyone" (OpenAI). You cannot delete what a staff member's personal account kept.
- You have a vendor you never vetted. The rules for insurance agencies, advisers and tax preparers all expect you to choose and oversee the companies that handle client information. A personal chatbot account is a vendor nobody chose. The sector risks page sets out the rules for each.
- You can't see it. If something goes wrong, you may not know which client's data went where.
IBM's 2025 breach study, of 600 organizations that had a breach, found that "One in five organizations reported a breach due to shadow AI", and that organizations with high levels of shadow AI "observed an average of $670,000 in higher breach costs than those with a low level or no shadow AI" (IBM). Those figures are not limited to small firms, but the cause is the same at any size.
How to tell if your team is already doing it
The simplest test is to ask, and to make it safe to answer. Tell staff that you are setting AI rules, that nobody is in trouble for past use, and that you want to know which tools they use and for what. You will learn more from that conversation than from any software.
Then look for:
- AI tools on firm-paid cards or expense claims.
- AI note-takers joining video calls.
- Browser extensions on work computers.
- AI features recently switched on in your email, document, CRM or practice software.
What to do about it
We suggest three steps:
- Write a short policy. One or two pages: which tools are approved, what client data may go in, and who to ask. Our AI acceptable use policy template is free to copy.
- Give staff an approved tool. If the only approved option is "nothing", staff will keep using their own. Pick a business plan whose terms on training, retention and access you have read. What makes an AI tool safe to approve compares the common options.
- Decide what blocking, if any, you need. Blocking can back up a policy, but a block on the office network or work laptops does not reach a staff member's own phone. See Should we just block ChatGPT at work?
Where to go next
- How do I stop employees pasting client data into ChatGPT? A step-by-step plan for a small firm.
- What should an AI acceptable use policy include? With a free template.
- Should we just block ChatGPT at work? What blocking does and does not do.
- What happens when client data leaks through ChatGPT? The public incidents, and what they teach.
- Can staff use their personal ChatGPT accounts for work? Personal plans versus business plans.
- AI privacy concerns at work The five AI data privacy concerns that matter for a small firm.
- Shadow AI risks by sector Insurance, wealth management, accounting and professional services.
- What makes an AI tool safe to approve for your staff? The questions to ask, and how the common plans answer them.