When the agent has to say it's an agent, what changes?

The EU's AI Act now treats disclosure as part of the media layer.

From 2 August 2026, Article 50 applies to certain AI systems. The European Commission says interactive AI systems must tell people when they are dealing with AI. Systems that generate synthetic audio, images, video, or text must make their outputs machine-readable and detectable as artificially generated or manipulated. Deployers must also disclose deepfakes and certain AI-generated text published on matters of public interest. 1

The handoff matters. A provider is responsible for how the system signals itself and marks what it generates. A deployer is responsible for how that system is used or published. An independent legal briefing says the rule reaches agentic systems that autonomously contact people, and that disclosure must be clear at first interaction rather than buried in terms or menus. 2

There is a meaningful editorial exception: public-interest AI text can avoid disclosure when it has undergone human review or editorial control and a natural or legal person holds editorial responsibility. Some standard editing is also outside the marking duty when the system does not substantially alter the input or its semantics. 3

My read: Agentic Media now has at least two provenance surfaces, the artifact and the actor. The label answers "is AI involved?" It does not yet answer who authorized the act, what changed, or who owns the final decision.

Discussion: What is the minimum disclosure you would want before an agent acts on someone's behalf?