A transparency notice is a statement that tells people how an organisation uses AI or their personal data, in terms they can understand and act on. For AI agents, it typically tells a customer they are dealing with an agent, what it does, what data it uses, and how to reach a person or challenge an outcome.
Why does a transparency notice matter for agents?
People are entitled to know when they are dealing with an AI system. The EU AI Act's Article 50 requires that people be told when they interact with an AI system, and that certain AI-generated content be marked; the detail is covered in Article 50 transparency for AI agents. Data protection law also requires organisations to explain how personal data is used, including in automated decisions.
The notice creates a promise. If it says "you can ask for a person to review any decision", someone has to be able to do that well. A customer who asks, and reaches a member of staff who cannot explain what the agent did, has been let down twice.
How does it work in practice?
A useful transparency notice for an agent covers:
- that the customer is interacting with an AI agent, stated up front;
- what the agent can and cannot do for them;
- what information it uses, in plain categories;
- which decisions a person reviews, and which the agent makes within limits;
- how to reach a person, and how to challenge an outcome.
Behind it, staff need to be able to deliver each promise: recognise when a customer wants a person, read the agent's record to explain a decision, and know when to reconsider. That is a training task, set out in training staff to tell people they are dealing with an AI agent.
What is an example?
An online retailer's chat agent opens each conversation with a short notice: it is an AI agent, it can track orders and start returns, a person reviews refunds over a set amount, and typing "person" reaches the team. A customer whose refund was held asks why. The service lead reads the record, sees the refund was held by the customer's own limit, explains this and approves it after checking. The notice set an expectation; the trained lead met it. Retailers face this daily; see what AI literacy means for online retailers.
How is it different from an adverse action notice?
A transparency notice is general and given up front: this is how we use AI. An adverse action notice is specific and given after a decision against a person, such as a credit refusal, stating the reasons. In the US, ECOA and Regulation B require specific reasons when credit is denied, including when a model is used. One explains the system; the other explains a decision.
Both rest on broader AI transparency: the principle that people affected by AI should be able to understand it.
Which terms are related?
- Privacy notice: the data protection counterpart, often combined with it.
- Automated decision: what the notice may need to describe.
- Decision literacy: the staff skill that makes the notice credible.
For the wider topic, read AI literacy and certification for people who work with agents.
Questions people ask
Where should a transparency notice appear?
At the point of interaction, before the customer relies on the agent: the start of a chat, the top of an email, the first message in an app. A link to a fuller policy can sit behind it.
Who should write it?
Legal and compliance for accuracy, with the agent's owner for what it actually does, and frontline staff to test whether customers will understand it.