The first ethical rule for an AI interface is honesty: the user must know when they interact with an automated system, when content is generated and when a choice is influenced by an algorithm. It is not legal pedantry — it is respect for decision-making autonomy.
Effective disclosure is contextual, not bureaucratic. A generic "AI" badge in the footer is not enough. Better: "This summary is automatically generated — verify the sources", "The price shown is personalised based on your profile", "You are talking to an AI assistant, not a human operator".
Tone matters. Fearful disclosure ("Warning: artificial intelligence!") creates unnecessary anxiety. Absent or hidden disclosure creates miscalibrated trust — the user treats AI output as human or expert truth. The designer seeks balance: clarity without alarmism.
Transparency also includes limits. Healthcare, legal or financial chatbots must communicate what they cannot do: "I am not a doctor", "I do not replace legal advice". Well-designed limits reduce harm from over-reliance — when the user depends excessively on an inadequate system.
Cases like Air Canada's chatbot (2024), which invented a ticket policy and the airline tried not to honour it, show what happens when disclosure and limits are absent. The interface gave authority to a hallucinated output. Designing for the worst case is not pessimism — it is due diligence.
Practical checklist: every AI feature in your product should answer — Does the user know there is AI? Do they know what it does? What it does not do? Can they disable it? Can they report errors? If an answer is missing, there is a design gap to fill.