Key principles
Introduction
Section titled “Introduction”🎯 Learning goals
- Understand data minimization and why it matters
- Know what purpose limitation means in practice
- Understand why you should always verify AI output about real people
- Know when and how to be transparent about AI involvement
Even when you’re using the right assistant for the right data, a few basic principles should guide how you work. These come from the same underlying logic as data protection law, and they also make practical sense.
Data minimization
Section titled “Data minimization”This is called data minimization. Before you paste something into an assistant, ask: does this task require me to include this person’s name? Their address? The details of their case history? In many cases, you can accomplish the same result with less.
Example: You want the assistant to help you structure a response letter about a housing benefit decision. You don’t need to include the applicant’s full medical history — the relevant decision and their name may be sufficient.
Purpose limitation
Section titled “Purpose limitation”This is called purpose limitation. Your organization’s assistants were built with specific use cases in mind. Using a case management assistant to, say, analyze patterns across many individuals, or to generate reports about groups of people, may take you well outside what it was set up to do.
When a new use idea feels meaningfully different from the assistant’s intended purpose — talk to your supervisor or data protection officer before proceeding.
AI output can be wrong
Section titled “AI output can be wrong”AI assistants can produce confident-sounding text that is inaccurate. This matters especially when the output concerns a real person. Always verify factual claims about individuals before acting on them or forwarding them.
Be transparent about AI involvement
Section titled “Be transparent about AI involvement”If an AI assistant contributed to a decision or a document that affects a person, that person is generally entitled to know. This doesn’t have to be complicated — a brief note that the work was AI-assisted, with a human having reviewed and taken responsibility for the final output, is often sufficient. Your organization may have specific guidelines on this.
Test your knowledge
4 questions · 100% correct to pass · Review your answers when done