Here is a question. Right now, this moment, do you know what data your employees are typing into ChatGPT?
I asked that question to a room of small business owners last year. Hands went up for "yes, I trust my team." Not a single hand for "yes, I actually know."
That is the gap. And it is a big one.
---
The Thing Nobody Talks About at the AI Hype Party
Everyone is cracking on about how AI is going to save your business twelve hours a week and do your marketing while you sleep. That is fine. Some of it is even true. But there is a conversation that is not happening loudly enough, and it is this: your team is probably already using AI tools, right now, with data they should not be putting in there.
I am not talking about rogue employees doing dodgy things. I am talking about well-meaning people, trying to move faster, doing what feels like sensible workflow shortcuts.
Sarah in accounts pastes a client's financial projections into ChatGPT so she can write a cleaner summary email. Your sales bloke copies a prospect's full contact record into Claude to help draft a personalised pitch. Your office manager feeds a staff performance review into an AI tool to get it polished up before the quarterly meeting.
None of them think they are doing anything wrong. They are just trying to get the job done.
But here is the thing. That data just left your building.
---
What Actually Happens When Data Goes Into a Public AI Tool
Let me be straight with you about what these tools actually are and are not.
ChatGPT, Claude, Gemini, and most of the popular AI assistants have two versions: consumer (free or cheap) and enterprise (more expensive, with proper data controls). The consumer versions are not enterprise-grade data vaults. Depending on the tool and its settings, the conversations your staff are having can be stored, reviewed by humans for safety and quality purposes, and in some cases used to improve the model.
OpenAI does let you turn off training data usage, but only if you know the setting exists and have actively changed it. Most people have not.
Now layer on top of that: in 2025, security researchers found more than 225,000 sets of ChatGPT credentials being sold on dark web markets, stolen by malware. Once someone has your login, they have access to everything your team ever typed into that account.
Still feeling relaxed about what Sarah put in there last Tuesday?
---
The Numbers Are Not Comforting
This is not me being dramatic. Here is what the research actually shows.
According to data from 2025, sensitive information makes up 34.8% of what employees type into ChatGPT. That number was 11% back in 2023. It is more than tripled in two years because more people are using these tools for more serious work tasks.
A separate survey found that 26% of organisations report that over 30% of what their employees feed into public AI tools is private or sensitive data. One in four businesses has a quarter of their staff's AI inputs being stuff that has no business leaving the company.
And for the small business owners in the room, here is the really sobering bit: only 9% of small companies monitor their AI systems for accuracy or misuse. Nine. Compared to 48% of larger organisations.
Which means if something goes wrong, you probably would not know until a client tells you. Or until someone's lawyer writes you a letter.
---
The Samsung Story (This Actually Happened)
This is not hypothetical.
Samsung engineers were using ChatGPT to help them with coding tasks back in 2023. In the process, they accidentally uploaded proprietary source code and internal meeting notes to the platform. When Samsung realised what had happened, they banned all external AI tools across the company immediately.
Samsung is a company with a full IT department, security teams, and policies. They still got caught out because employees were just trying to work faster and the tool was right there.
If it can happen to Samsung, it can happen to your five-person accountancy firm or your recruitment agency or your e-commerce shop. The difference is Samsung could absorb the reputational and legal risk. You might not be able to.
---
What Your Business Actually Needs to Worry About
Not everything carries the same level of risk. Here is how I would think about it.
High risk, never goes in a public AI tool:
- Customer names combined with financial or health information
- Employee personal details, salary data, or performance reviews
- Legal documents, contract terms, or anything under an NDA
- Passwords, API keys, or anything authentication-related
- Patient information if you are anywhere near healthcare
Moderate risk, use your judgement:
- Internal meeting notes (redact anything client-specific first)
- Draft proposals (remove specific pricing or confidential client details)
- Customer communications (anonymise where you can)
Low risk, go for it:
- Brainstorming content ideas
- Rewriting your own marketing copy that is already public
- Summarising your own website content
- Writing job descriptions for advertised roles
The rule I use with clients: if you would not type it into a Facebook status or paste it into an email to someone you have never met, it should not go into a free AI tool without thinking twice about it.
---
The Uncomfortable Bit About GDPR (Sorry)
If your business is in the UK or serves UK or EU customers, you have got GDPR obligations. One of those obligations is that personal data should only be processed in ways your customers have consented to or that you have a legitimate legal basis for.
"I pasted their contact details into ChatGPT because it was faster" is not a GDPR-compliant processing activity.
Now, is the ICO going to come knocking because your sales team used Claude to write a pitch email? Probably not. But if there is a breach, if customer data ends up somewhere it should not, and the ICO starts asking questions, "we had no AI policy and no controls" is not going to go well for you.
The good news is the fix is not complicated. You do not need an enterprise AI platform costing thousands a month. You need a simple policy and about thirty minutes of team training.
---
The Actual Fix (It is Simpler Than You Think)
Step one: find out what your team is already using. Just ask them. Most people will tell you. You might be surprised.
Step two: decide on one approved tool for work tasks. OpenAI offers a Team plan at around $30 per user per month. That version does not use your conversations to train the model by default. Claude has business tiers too. Pick one and pay for it. The cost is trivial compared to the risk.
Step three: write a one-page AI use policy. It does not need to be legal jargon. It needs to say: here are the tools we use, here is what you can put in them, here is what you cannot. Specifically list the data types that never go in. Print it out. Put it somewhere people will see it.
Step four: do a fifteen-minute team briefing. Show people the Samsung story. Show them the 34.8% stat. Make it real. People are not stupid. When they understand the actual risk, they change their behaviour.
Step five: revisit it every six months. AI tools change fast. Your policy needs to keep up.
That is it. Four steps, one afternoon, done.
---
The Bit I Actually Want You to Take Away
AI tools are genuinely useful. I use them every day. My clients use them. The goal here is not to scare you off AI. The goal is to make sure you are using it with your eyes open.
The businesses that will win with AI are not the ones who let their team use every free tool they can find without thinking about it. They are the ones who use it deliberately. Who know what is going in and what is coming out. Who have made considered choices about which tools to trust with which data.
Right now, 91% of small businesses have no monitoring on what their AI tools are doing. Most do not have a policy. Most do not know which tools their staff are using.
You can be in the 9% just by reading this and doing something about it.
That is not a massive ask. It is one afternoon. It might save you a client, a fine, or a very uncomfortable conversation.
---
If you want to talk this through with people who are actually doing it, come join us in Business Without the Bullsh*t on Facebook. No gurus, no fluff, just real conversations with other small business owners working it out.
---
About Steven Tann: Steven is "The Bloke Who Fixes Your Tech Stack." With over 10 years in the trenches helping more than 7,000 small and medium businesses, he cuts through the guru fluff and builds AI, marketing and automation systems that actually work for real business owners. No jargon. No upsells. Just sorted. Find out more at steventann.com.