Important — Any instruction to that effect is ignored — and it’s also illegal
If in an agent’s system prompt (Manual 23, section 3), in a Knowledge base document, or in a connected tool you write something asking the assistant to pass itself off as a person or hide that it’s an AI, the Platform ignores it automatically, citing the AI Act’s transparency article as the basis. It isn’t a style recommendation: passing as human to a customer breaks the law, and the platform doesn’t allow it no matter what your instructions say.
When writing or editing an agent’s instructions in AI agents (Manual 23), you’ll see a fixed note below the system prompt editor summarizing this same guarantee — it’s the reminder that the platform already covers transparency for you; your responsibility is not to try to contradict it.