Is our data safe if you use AI?

It is, when the system is designed for it from the start. That means deciding what personal data the AI can see, where it’s processed, under whose terms, and who checks what comes out. We settle those questions before we write any code, not after a customer asks.

We’re engineers, not lawyers. What follows is how we design AI systems to stay within the rules, using the guidance the regulator publishes. It’s general information, not legal advice.

Where your data actually goes

There are two routes, and the choice shapes everything else.

Through a model provider. Your data is sent with each request to a model run by a company like OpenAI, Anthropic or Microsoft. It isn’t trained into the model when the system looks things up as it answers. The provider still processes it, so its terms matter.

Before anything is connected, check four things in those terms:

• whether your data is used to train its models, and whether that’s off by default
• where the data is processed and stored, and how transfers abroad are covered
• how long it keeps what goes in and what comes out
• what security certifications it holds, and what happens after a breach

The business plans from the main vendors say inputs aren’t used for training by default. Free personal accounts carry different terms. So staff pasting work into personal accounts is the bigger risk, and paying for proper seats is partly a data protection decision. Check each vendor’s current terms yourself, because they change.

Kept in-house. Some clients don’t want sensitive data reaching any model vendor. That’s a reasonable position with a design answer. An open model can run on servers you control, or on the device itself. It costs more to set up and look after. A cloud host is still a processor, so choose a UK region to avoid international transfers.

What UK GDPR asks of you

UK GDPR has no separate rulebook for AI. The same principles apply as for any use of personal data, and AI just makes some of them harder.

• You stay the controller. If you decide what the AI is for and what goes in, the duties are yours. A vendor following your instructions is generally a processor.
• You need a lawful basis. Write down which one applies, and tell people in your privacy notice that AI is involved.
• Expect to need a DPIA. That’s a data protection impact assessment. The ICO says most AI use involves high-risk processing. If you decide yours isn’t, it expects you to record why.
• Automated decisions need safeguards. Since 5 February 2026, people must be able to challenge a significant decision made solely by AI, and ask for a person to step in.

The ICO says its AI guidance is under review after the Data (Use and Access) Act 2025. Check its current version before relying on any detail. Our guide to AI, GDPR and governance cites each source.

How we design for it

The risk we design against most isn’t a dramatic hack. It’s an assistant that can see too much.

• It sees only what it needs. An assistant over your whole shared drive can hand a payroll file to anyone who asks. We limit it to what the person asking may see.
• Outside content can’t give orders. A document or email can contain text written to steer the AI. We don’t let it act on outside content unchecked.
• Logs are treated as data. Prompts and answers often hold personal details, so logs get access controls and a retention period.
• A person checks what matters. We agree which outputs a person signs off before they reach a customer or a payslip.

Our infrastructure + security service covers how the rest is locked down.

What we hold, and what we prepare you for

Contic holds Cyber Essentials Plus and is registered with the ICO. We manage our own compliance in Secureframe, aligned to SOC 2 and ISO 27001.

For your product, we work towards readiness for ISO 27001, SOC 2 Type II, HIPAA and GDPR. We get you ready for the audit. The certificate itself comes from an independent auditor, not from us.

Vault shows what that looks like in practice. Reports there are encrypted on the person’s own phone, with keys held by them and their employer. Even Vault can’t read them. We took the platform through SOC 2 and ISO 27001, the certifications its enterprise buyers asked for.

When you need a lawyer, not us

If you need a legal opinion on your own processing, that’s a job for a data protection lawyer or your DPO. Where we help is designing the system around the rules: what it sees, what it logs, where it runs and when a person steps in.

If you’re still deciding what AI should do for you, start with can you add AI to our existing product?. Who can reach your code and data once it’s built is covered in do I get access to the code, CMS and analytics?