Protecting your privacy when using AI tools
What you type into an AI chatbot is visible to the company that provides it, and queries stored online may be hacked, leaked or accidentally made publicly accessible (UK NCSC).
What regulators and cyber-security agencies advise
- Keep sensitive information out of public chatbots. The UK National Cyber Security Centre recommends not including sensitive information in queries to public AI language models, and not submitting queries that would lead to issues if they were made public (NCSC).
- Read the terms of use and privacy policy first. The NCSC says queries are stored and will almost certainly be used for developing the service at some point, so the terms of use and privacy policy need to be understood before asking sensitive questions (NCSC).
- Leave out personal details. Spain's data protection agency advises not sharing personal data with AI tools, such as your full name, address, phone number, identity document number or images of people, or sensitive information such as medical or financial details, and suggests describing a fictional case instead (AEPD).
- Do not use images of other people to generate new content, especially images of children; the agency warns that this could breach data protection law and even be a crime (AEPD).
- At work, follow your organisation's rules on information and security, and do not include confidential information about the organisation, its staff or its clients (AEPD).
- Chat history settings. Guidance for judges and their support staff in England and Wales tells them to disable the chat history in public AI chatbots if this option is available, as it should prevent their data from being used to train the chatbot, but warns that even with history turned off it should be assumed that data entered is being disclosed (judicial guidance).
- Be careful with AI tools that process content written by others. Hidden text in a document, such as a CV, can end up being treated by an AI system as an instruction. The NCSC calls this indirect prompt injection and says such attacks may never be totally mitigated (NCSC).
- Companies must keep their privacy promises. The US Federal Trade Commission has warned "model-as-a-service" companies, which develop and host AI models for others to use, that if they fail to abide by their privacy commitments to their users and customers, including promises not to use customer data for secret purposes such as training their models, they may be liable under the laws it enforces (FTC).
- If you think your data has been mishandled, in most cases raise your concern with the organisation first. In the UK, if you are unhappy with its response, you can then complain to the Information Commissioner's Office (ICO).
Sources
- ChatGPT and large language models: what's the risk?, UK National Cyber Security Centre, 14 March 2023.
- Prompt injection is not SQL injection (it may be worse), UK National Cyber Security Centre, 8 December 2025.
- La AEPD publica un decálogo con recomendaciones para proteger la privacidad al usar herramientas de IA, Spanish Data Protection Agency (AEPD), 27 January 2026.
- Artificial Intelligence (AI) – Judicial Guidance (October 2025), Courts and Tribunals Judiciary, 31 October 2025.
- AI Companies: Uphold Your Privacy and Confidentiality Commitments, US Federal Trade Commission.
- Check if you can complain, UK Information Commissioner's Office.