Data, privacy and confidentiality
The fastest way to get into trouble with AI is to paste the wrong thing into the wrong tool. This module shows you how to keep personal and confidential information safe.
Learning objectives
- Explain the difference between consumer and enterprise AI tools for data handling
- Classify information using a green–amber–red approach
- Apply basic data protection principles when using AI
- Recognise confidentiality and intellectual property risks
- Build safe everyday habits
4.1 Where does your data go?
When you type into an AI tool, your text (and any files you upload) is sent to the provider's servers to be processed. What happens next depends heavily on which version of the tool you're using.
Consumer / personal accounts
Free or personal plans may, depending on the provider and your settings, keep conversations and use them to improve models. You're bound by consumer terms, not your employer's contract.
Enterprise / business accounts
Business plans typically come with a contract that sets out data handling, retention, security and — usually — a commitment not to train models on your data. Your IT and legal teams have reviewed it.
AWS's Generative AI Security Scoping Matrix makes the same distinction for organisations: using a public consumer app (scope 1) gives you the least control over data, while an enterprise app under a business agreement (scope 2) gives you more. Settings and terms change, so always check your organisation's current guidance.
4.2 A traffic-light approach to data
Many organisations use a simple classification like this. Your own policy takes precedence.
| Type of information | Examples | Typical rule | |
|---|---|---|---|
| Green | Public or non-sensitive | Published web content, generic questions, your own rough ideas | OK in approved tools |
| Amber | Internal | Internal process docs, draft plans, non-personal operational data | Approved enterprise tools only |
| Red | Confidential, personal or regulated | Customer records, HR files, health data, passwords and keys, unreleased financials, client confidential material | Not without specific approval, safeguards and a lawful basis |
4.3 Personal data and UK GDPR
If you put information about identifiable people into an AI tool, data protection law applies — in the UK, UK GDPR and the Data Protection Act 2018; in the EU, the GDPR. In practice that means:
- Only use what you need (data minimisation). Can you remove names or replace them with placeholders?
- Have a lawful basis and use the data in ways people would reasonably expect.
- Be careful with special category data such as health, ethnicity or religion.
- Be cautious about automated decisions that significantly affect people — such as hiring or credit — which carry extra legal protections.
The UK Information Commissioner's Office (ICO) publishes detailed guidance on AI and data protection — see the links below.
4.4 Confidentiality and intellectual property
- Client and commercial confidentiality: contracts and professional rules (for lawyers, accountants, doctors) may forbid sharing information with third-party tools.
- Your organisation's IP: source code, product plans and trade secrets need the same care.
- Others' IP in outputs: AI can produce text or images that closely resemble existing works. Check before publishing, and follow your organisation's rules on using AI-generated content commercially.
4.5 Safe habits
- Use approved, enterprise accounts for work
- Strip out names, contact details and identifiers when you can — "Customer A", "Employee B"
- Never paste passwords, API keys or access tokens
- Check your settings for chat history and model training
- Be careful with uploads — a spreadsheet may contain hidden tabs or columns of personal data
- When in doubt, ask your data protection or information security team first
"Summarise this complaint from Jane Smith (jane.smith@email.com, account 12345678) about her mortgage arrears and draft a reply."
Show a safer version
"Summarise this customer complaint about mortgage arrears and draft a reply. [Complaint text with name, email and account number removed]". Use an approved enterprise tool, and only if your policy allows customer correspondence there. Add personal details back yourself before sending.
Key takeaways
- Enterprise AI plans usually offer contractual data protections that consumer plans don't.
- Classify information before sharing: public, internal, or confidential/personal/regulated.
- Data protection law applies to personal data you put into AI — minimise and anonymise.
- Never paste secrets; check uploads; ask when unsure.