EU AI Act Art. 50(4): AI text on public-interest topics and scripts

Article 50(4) asks deployers to disclose AI text on public-interest matters unless a human reviewed it editorially. What it means for video scripts.

4 min readSume
All posts

Article 50(4) of the EU AI Act says deployers who publish AI-generated text to inform the public on matters of public interest must disclose that, unless a human has reviewed or edited it editorially. If your video script was written by a language model and the video informs people about a public-interest topic, a documented human edit is the exemption to look at.

What Article 50(4) says

The text of Article 50, read on 2026-10-03, sets deployer duties in paragraph 4. This post covers the text row; the first two rows are about deepfakes and creative works.

Article 50(4) in short (read 2026-10-03)
CaseDuty on the deployer
Deepfake (artificially generated or manipulated image, audio or video)Disclose that it is artificially generated or manipulated
Evidently artistic, creative, satirical or fictional workLimited disclosure that does not hamper display or enjoyment of the work
AI-generated text published to inform the public on public-interest mattersDisclose, unless human review or editorial control applies

Where video scripts fall

A narrated explainer on a health, civic or financial topic is the kind of video this paragraph is about when the script text came from a model and the video is published to inform the public. A product demo or entertainment clip is a different case. The article does not turn on the format of the video, only on what the text is for and who edited it.

This is not legal advice, and whether a given video counts as informing the public on a matter of public interest is for you and your counsel to decide, not for a generator or a blog post.

Make the human review provable

The exemption turns on human review or editorial control, so the useful thing is a record that someone actually reviewed the text. Store the final script, the reviewer, and the date next to the video it produced.

Sume's video docs do not describe a field for reviewer records, so keep the record in your own store, keyed by the job id that Sume returns. A reviewer id and a hash of the approved script are enough to tie the proof to the clip.

import hashlib
import json

script = "Approved script text goes here."
digest = hashlib.sha256(script.encode("utf-8")).hexdigest()

# Store this next to the job id returned when you submit the video.
record = {
    "job_id": "job_...",
    "script_sha256": digest,
    "reviewed_by": "editor-17",
    "reviewed_on": "2026-10-04",
}
print(json.dumps(record, indent=2))

Checklist

Decide per video whether the script informs the public on a public-interest matter.

If yes, either have a person review and edit it editorially and keep the record, or disclose that the text is AI-generated.

Treat the visuals separately: a realistic synthetic person or event can be a deepfake under the first paragraph regardless of who wrote the script.

Sources

Related posts

More in Use cases

All Use cases posts

Written by Sume