AI and Data Security: How to Use AI Safely with Your Company's Data
How to use AI with company data without losing control: the three main risks, a safe setup, GDPR points and a checklist you can use straight away.
Table of contents
Contents
Short answer: you can use AI with your company's data without giving up control, but it requires choosing the right tools and setup. It comes down to three things: that data doesn't leave your control, that it isn't used to train other people's models, and that only the right people have access. With a paid business solution (or your own instance), a data processing agreement and control over access, you can get the benefit without breaching GDPR. Below are the risks, the safe setup and a checklist you can use straight away.
Why it's urgent, and why most people get it wrong
Most employees already use AI. Often it's the free, public versions, where they copy customer data, contracts, emails or spreadsheets into a text box without thinking about where it ends up. That is where the problem arises: not in the technology itself, but in how it is used.
For a business, the question isn't "may we use AI?": it's "on what terms?". Once you have answered that, AI becomes one of the most powerful tools you can give your employees. Without an answer, it is an open door to your data.
The three big risks
1. Data leaves the building
When an employee pastes text into an AI service, it is by default sent to an external provider's servers, often outside the EU. You no longer have full control over where the data sits, who can access it, or when it is deleted. For personal data and trade secrets, that is the central risk.
2. Your data is used to train models
Several free AI services reserve the right to use what you enter to improve and train their models. In practice that means your internal documents can become part of a system you don't own and, in the worst case, turn up in answers to other users. Paid business agreements typically have the opposite starting point (training switched off by default), but you need to confirm that actively, not assume it.
3. Lack of access management
Without management, you don't know who is using AI for what, which data has been sent out, or whether a former employee still has access. Shadow AI (tools adopted without IT's knowledge) is a real risk, because it removes any overview. Security without logging is, in practice, no security.
How to use AI safely
The good news: all three risks can be managed. Here are the five measures, in order of priority.
Paid business tiers with training opt-out
The first and cheapest step: switch from free to business editions. Reputable providers offer business or enterprise plans where your data is by default not used to train models, and where you can enter into a formal data processing agreement. Confirm it in writing: read the terms, ask directly, and get it on paper.
Your own instance or fair source: your data stays yours
If you want to bring control fully in-house, a dedicated instance or a fair-source setup is the answer. Here the AI runs in an environment you control, and your data stays with you instead of flowing out to a third party's shared platform. That is the direction we are building towards in Agent360 OS, which is on its way and not yet released. For companies with sensitive data, customers in regulated industries or strict internal requirements, it is often the difference between "we can't" and "we can".
Encryption
Data must be encrypted both when it is stored and when it is sent over the network. It's hygiene, but it is crucial, and something you should be able to get confirmed by any reputable supplier.
Access control and roles
Not everyone should be able to do everything. Set up roles so that access matches the job function, and make sure access is removed when an employee leaves or changes role. Combine this with logging, so you can see who has used what, both for security and to be able to document it afterwards.
Data processing agreement (DPA)
If you use an external provider to process personal data, GDPR requires a data processing agreement between you (the data controller) and the provider (the data processor). Without it, you lack the legal basis. It isn't negotiable, it's a requirement.
GDPR considerations you need to have under control
AI doesn't change the basic GDPR principles, it just makes them more relevant. The most important:
- Lawful basis: You need a valid basis for processing personal data in the AI, just as with any other processing.
- Data minimisation: Send only the data that is genuinely necessary. There is seldom a reason to feed an AI the entire customer register to solve one task.
- Purpose limitation: Data collected for one purpose may not simply be reused freely for another.
- Data subjects' rights: People still have the right of access, erasure and objection, even when their data has been through an AI system.
- Transfers to third countries: If data is processed outside the EU/EEA, the basis for the transfer must be in place. Keep data in the EU where you can.
- Accountability: You must be able to document that you comply with the rules. This is where an audit trail (a log of what has happened) becomes your best friend.
Please note: this is a general overview, not legal advice. If you are in doubt about your specific situation, involve your DPO or a data protection lawyer.
What to ask a supplier
Before you sign, get clear, written answers to these questions. If a supplier can't answer clearly, that is an answer in itself.
- Do you use our data to train your models? (The answer must be no; confirm in writing.)
- Where is our data stored and processed geographically?
- Do you offer a data processing agreement, and what does it say?
- Is data encrypted in storage and in transit?
- Can we manage access at role and user level?
- Do we get a log/audit trail of usage?
- How long is data kept, and how is it deleted?
- Can we get a dedicated instance, or do we share a platform with other customers?
- Who at your company can in practice access our data, and under what conditions?
- How do you handle security incidents and notification?
Checklist: safe AI use in practice
- [ ] Switch from free to paid business tiers with training switched off
- [ ] Consider your own instance / fair source for sensitive data
- [ ] Data processing agreement (DPA) in place with every provider that touches personal data
- [ ] Encryption confirmed, both stored and in transit
- [ ] Access control at role level, and access removed on departure
- [ ] Audit trail / logging of who uses what
- [ ] Data kept in the EU where possible
- [ ] A clear internal policy on what may and may not be entered
- [ ] Data minimisation, only necessary data is sent in
- [ ] Employees are trained, so shadow AI doesn't arise
How Agent360 helps
At Agent360, we help businesses adopt AI without giving up control of their data. We provide advice on the right setup, and we are building Agent360 OS, where each customer should be able to have their own instance with their own data; it is on its way and not yet released. We work to recognised security principles and help you get the right agreements and access management in place, so that AI becomes an advantage, not a risk.
Getting started safely
AI can lift your business markedly when it is set up correctly. If you are in doubt about how to use AI safely with your own data, let's have a chat about your specific situation.
Book an AI consultation with Agent360, and we will go through your needs, your data and the setup that gives you the benefit without the risk.
Contents
What takes up your time?
Write down the task that takes up the most time. We reply whether it is worth building, and we say so if it is not.
Write what takes up your timeRelated articles

MFA: multi-factor authentication for your business – how to get started
Learn how to implement MFA securely in your organisation. Guidance on methods, costs and best practices for protecting company data.

English-language invoices: requirements for invoices to foreign customers
Invoice in English to foreign customers legally. Danish regulations, EU and international VAT rules, zero-rating, reverse charge, and automation.

Drawing programme (CAD): 2D, 3D, licences and file formats
What is CAD? Learn about 2D and 3D design programmes, file formats, licensing models, and how to choose the right tool for technical drawings.