Using AI responsibly protects your customers, your reputation and you: legally and commercially, not just morally.

Fair, transparent, privacy-respecting AI keeps customers. Source: OECD AI Principles (2024).

Bias means an AI treats some groups unfairly. It usually comes from the data, not from bad intentions.

• It starts in the data. AI learns from human-made examples. If they reflect past unfairness, the AI can learn and repeat it.
• A simple example. A hiring AI trained mostly on one group's CVs may quietly score similar candidates higher, disadvantaging others.
• Why it's dangerous. AI is fast and confident, so biased decisions can be made at scale before anyone notices.

AI learns from the past, so it can repeat the past's unfairness.

Bias wears several disguises. Recognising them is the first step to catching them.

• Data bias. The training data over- or under-represents certain groups, skewing results against them.
• Sampling bias. The data was collected in a way that misses part of the real population.
• Historical bias. The data reflects past inequalities, which the AI then treats as 'normal'.
• Automation bias. People trust the machine too much and stop questioning it. A human failing, not the AI's.

Ask 'who might this treat unfairly?' for every AI output you rely on.

The EU AI Act sorts AI by risk: the higher the risk, the stricter the rules. Place your project before you build.

Source: EU AI Act, Regulation (EU) 2024/1689.

The duty you'll meet most often. In plain terms: be honest that AI is involved. It applies from 2 August 2026.

• Say it's AI. Chatbots and AI assistants must make clear that users are talking to a machine, unless it is obvious.
• Label AI-made content. Deepfakes, and AI text published to inform the public, must be disclosed as AI-made. AI tools must mark their output.
• Why it matters. Transparency lets people judge AI output for themselves. It is the foundation of trustworthy AI.

Source: EU AI Act, Art. 50 (transparency); applies from 2 August 2026.

Your four duties

• Be transparent. Tell people clearly when they deal with AI or AI-generated content.
• Check for bias. Test that your AI treats different groups fairly, not just on average.
• Protect data. Collect only what you need and store personal data securely (GDPR).
• Keep a human in the loop. A person reviews and can override important AI decisions before they take effect.

If your AI touches personal data, the GDPR applies. Collect the least you can: what you don't hold can't leak.

• Have a lawful reason. You need a valid basis to use someone's data, usually clear consent or a genuine business need.
• Collect only what you need. 'Data minimisation' means gathering the minimum required, not everything you could.
• Respect people's rights. People can ask what data you hold, correct it, or have it deleted. Be ready to honour that.
• Decisions by machine. If AI alone decides something important, like a loan, people can ask for a human to review it (GDPR Art. 22).

Reference: GDPR, Regulation (EU) 2016/679.

An ethics audit is a simple, repeatable check. Five steps make it manageable.

1. Set the scope. Pick one AI tool, such as your chatbot, and define what it does and who it affects.
2. Check the data. Ask what data it uses and whether any group is missing or misrepresented.
3. Test for bias. Try inputs from different groups and look for unfair differences in the output.
4. Check transparency. Confirm it discloses that it's AI and protects personal data properly.
5. Log gaps and fix. Write down what's wrong, plan changes, and re-check after fixing.

In EX 4.4 you apply a 20-point checklist to your EX 4.1 chatbot.

The line between the two is mostly honesty, fairness and care.

Aim for the left column in everything you build.