Your staff are probably using AI already. Someone has pasted a customer email into a chatbot to get a better reply, or dropped a spreadsheet in to summarize it. So the practical question for an owner is how to keep business data private while people use it. This guide covers where AI data privacy for business actually breaks down, what the law expects of you, and a checklist you can work through with your team.
This is general information for business owners, not legal advice. For decisions about your own obligations, speak to a qualified data protection adviser.
How business data leaks into AI tools
Most leaks start with ordinary people trying to do their jobs faster, using whatever tool is nearest.
The case most people remember is Samsung. In 2023, after engineers uploaded internal source code to ChatGPT, the company banned staff from using generative AI tools on company devices. Its concern was simple: once data sits on an outside provider's servers, it is hard to get back or delete.
Samsung is not unusual. Cisco's 2024 Data Privacy Benchmark Study, a survey of privacy and security professionals, found that more than one in four organizations had banned generative AI at least for a time, and many admitted that employees had entered non-public company information and employee or customer details into these tools.
The usual routes look like this:
Free consumer accounts. Personal chatbot accounts often run on terms that allow the provider to use conversations to improve its models, unless the user switches that off.
Browser extensions and plug-ins. Handy add-ons that read everything on screen, including your CRM and email.
Meeting recorders. Tools that join calls, transcribe them and store the transcript somewhere you never chose.
Shadow projects. A team signs up for an AI service on a credit card, uploads a few thousand customer records and nobody in finance or IT knows.
Bans rarely work for long
A blanket ban feels safe, but staff who rely on AI usually just switch to personal devices and accounts, and then you cannot see what they share. A better approach is to give staff an approved way to use AI and make the unapproved ways unnecessary.
That means deciding three things: which tools are approved, which data may go into them, and where the data is processed and stored.
What the law expects: GDPR and the EU AI Act in plain terms
If you handle personal data of people in the EU or UK, data protection law applies to AI use just as it does to any other software. A few points matter most.
The AI provider is usually your processor (or more)
When an outside service handles personal data on your behalf, Article 28 of the GDPR requires a written contract with that processor.
Among other things, it must act only on your documented instructions, keep the data secure, get your authorization before using sub-processors, and delete or return the data when the service ends. In practice this is the data processing agreement, or DPA. If an AI vendor cannot give you one, that tells you a lot.
The roles are not always that tidy with AI. The UK Information Commissioner's Office has said that a contract does not by itself decide whether an organization is a controller, joint controller or processor; what happens to the data in practice does. If a provider uses your data to train its own models, it is making its own decisions about that data, and the picture changes.
The EU AI Act adds its own duties
The EU AI Act entered into force on 1 August 2024 and applies in stages. Since 2 February 2025, certain AI practices are banned and organizations that deploy AI systems are expected to make sure their staff have a sufficient level of AI literacy.
Rules for providers of general-purpose AI models have applied since 2 August 2025, and most remaining obligations are scheduled to follow from August 2026. For a typical mid-size firm using AI for office work, the near-term tasks are staff training and knowing which AI systems you use and for what.
The US has no single federal equivalent, but state privacy laws, sector rules such as HIPAA for health data, and the terms of your own customer contracts can all restrict where data goes. Check your customer contracts in particular. Many include confidentiality clauses that a careless upload would breach.
AI data privacy for business: an eight-point checklist
Work through these with whoever runs IT and whoever owns data protection. Most firms can get through the list in a few meetings.
1. Know what is in use today
Ask each department which AI tools they use, on which accounts, with what data. Expect surprises. You cannot protect what you have not listed.
2. Sort your data into three bands
Open: public information, marketing copy, published prices.
Internal: processes, internal reports, non-sensitive figures.
Restricted: personal data, customer contracts, health or financial records, source code, unreleased plans.
Then set a simple rule for each band. Restricted data should only ever go into tools you have contracted for.
3. Use business terms, not consumer terms
The major AI providers offer business and API plans with different data terms from their free consumer apps, typically including a commitment not to train on your data and a DPA. Read the terms of the exact plan you buy, not the marketing page.
4. Decide where data is processed
For many uses, a reputable cloud AI service under a DPA is enough. For regulated or highly sensitive work, you can run AI inside your own cloud account, in a chosen region, or use open models hosted on your own infrastructure so documents never leave it.
5. Give each person only what their role allows
An AI assistant connected to your file shares can surface anything it can read. Make sure it respects the same permissions as the person asking, so a sales rep cannot ask it for payroll data.
6. Keep a person in the loop for anything that goes out
Privacy is also about what AI sends. Emails to customers, changes to records and payments should wait for a named person to approve them.
7. Log what the AI does
Keep a record of what was asked, which documents were used and what came back. If a customer or regulator asks, you can answer.
8. Train your people
A one-page policy and a 30-minute session cover most of it: which tools are approved, what never goes into them and who to ask. It also goes a long way toward the AI literacy the EU AI Act expects.
Questions to ask any AI vendor
Will you use our data, prompts or outputs to train any model?
Where is our data stored and processed, and can we choose the region?
How long do you keep it, and can we have it deleted?
Which sub-processors do you use, and will you tell us before adding new ones?
Will you sign a DPA and, where needed, an NDA before we share anything?
Who owns the prompts, configurations and logs if we stop working with you?
If you would like an independent view of a vendor's answers, or help deciding between a cloud service and a self-hosted setup, that is the kind of question our software and AI consulting is built for. We recommend; you decide.
How we handle it when we build AI for clients
When we build AI agents for a business, the defaults are the ones on this checklist. We sign an NDA, MSA and DPA before we see data.
We use business-grade AI services with training switched off, keep documents in the client's own cloud account wherever their policy requires it, and can run open models on the client's own infrastructure for regulated work. Each person sees only what their role allows, and nothing is sent, changed or paid without a named person approving it.
Frequently asked questions
Is it safe to use ChatGPT or similar tools with company data?
It depends on the plan and the data. A free personal account is not the right place for customer records or contracts. A business plan with a DPA and training switched off can be suitable for many tasks. Restricted data may need a setup that runs inside your own systems.
Do we need a DPA with an AI provider?
If the provider processes personal data on your behalf and GDPR or UK GDPR applies to you, Article 28 requires a written contract with the terms a DPA covers. Ask for it before any personal data goes in.
Does the EU AI Act apply to a company outside the EU?
It can. The Act reaches organizations outside the EU in some cases, for example where the output of an AI system is used in the EU. If you sell into Europe, take advice on how it applies to you.
Can AI run without our data leaving our systems?
Yes. Open models can run on servers or cloud accounts you control, so documents and prompts stay inside your environment. It costs more to set up than a hosted service, so it is usually kept for the most sensitive work.
A sensible next step for AI data privacy
If you are not sure which AI tools your teams already use, or what data has gone into them, that is the place to begin. Our free 5-day AI readiness audit looks at one department and tells you where AI fits and how to keep the data where it belongs. Or book a 30-minute call and talk it through with a founder.
