AI-written script on a public-interest topic: Article 50 text label

Only if the text meets all of the Commission's criteria. The Transparency obligations under Article 50 of the AI Act page says deployers must clearly label AI-generated or manipulated text published to inform the public on matters of public interest, unless it went through human review or editorial control. That page does not say whether burned-in video captions or a spoken script count as "published text," so for video the question stays open. This reports the page and is not legal advice.

What are the three criteria for the text?

The text must be published, informative to the public, and on matters of public interest. The page lists examples: politics and democratic processes, public administration and services, justice and law enforcement, fundamental rights, public security, public health, environmental protection, consumer safety, and economic, financial, political, scientific or cultural developments that may be a relevant subject of public debate.

What counts as human review or editorial control?

Human review is the deliberate examination of the substance by one or more people with relevant knowledge and professional judgement, such as peer review or professional validation chains. Editorial control is control exercised in practice by a responsible editorial entity, such as an editor-in-chief, with authority to approve, alter or reject the substance, including fact-checking and checking sources.

The page says superficial, solely formal or procedural checks, for example spell-checking or grammar correction, do not count. Editorial responsibility means a person holds ultimate legal responsibility for publishing the content.

Where does a Sume video script fit?

Sume's Video captions accepts script_text or authored cues and burns them onto a video. The text is whatever you pass in, so any review of it happens before you call the endpoint. If a caption track states facts on a public-interest topic, the page's test is about substantive review of those facts, not about the tool that rendered them. Keep a record of who reviewed what, alongside the job id from Jobs and results.