AI Compass
Compass

Media and publishing

Research, production and labelling: what holds in a newsroom, what the AI Act requires by way of transparency, and where copyright draws the line.

·2 min read·By Redaktion KI-Kompass
DETAIL
4 sections

What this is about

TaskClassification
Preparing research and collecting sourcesunproblematic, with review
Transcribing interviewsunproblematic
Translation and subtitlesunproblematic, with review
Summarising long documentsunproblematic, with review
Drafting headlines and teasersunproblematic
Generating or altering imageslimited, labelling duty
Recreating voices or faceshigh, personality rights

What holds in the newsroom

  1. 01

    Transcription

    The most immediate gain. An hour of interview is written out in minutes, and correcting it takes longer than the tool needs.

  2. 02

    Document review

    Searching large bundles for keywords and anomalies. Every find is checked in the original.

  3. 03

    Preparation

    The same story for different channels and lengths. No substitute for editorial selection.

  4. 04

    Archive work

    Tagging, linking and searching your own holdings.

Labelling in practice

  • One house rule on what gets labelled, settled before the first output is published.
  • Labelling on the content, not in the imprint. It has to sit where the content is seen.
  • Machine-readable marking where the format supports it, in addition to the visible one.
  • For images that could show real people or events, label when in doubt.

Art. 50 AI Act, applied

ContentDutyWho carries it
Interaction with a systemDisclose that it is not a personDeployer
Artificially generated image, audio or videoMachine-readable markingProvider
Deepfake with real people or eventsDisclose that the content is generatedDeployer
Text on matters of public interestDisclose where not editorially reviewedDeployer

The last row contains the exemption that relieves newsrooms: where a person takes editorial responsibility and checks the content, the disclosure duty for text falls away. That responsibility has to be exercised in fact, not merely asserted.

  • Training data. Text and data mining is permitted under the DSM Directive where no effective reservation of use is declared. A reservation on your own offering has to be machine-readable to take effect.
  • Generated works. Without a human creative contribution, no copyright. Anyone needing protection must be able to evidence their own contribution.
  • Archive rights. A right to publish is not a right to pass material to a model provider. Existing contracts usually say nothing, and silence is not permission.
  • Press publishers' right. Press publications are separately protected. That holds against systems that summarise and deliver content too.

See Copyright and AI.

Personality rights

The voice, face and gestures of an identifiable person are protected even where the result is entirely generated. Consent is purpose-bound: agreeing to a recording is not agreeing to a recreation.

Entry sheet: media and publishing

FREE ACCOUNT

Entry sheet for media and publishing

This sector's tasks by risk class, the points to settle beforehand, and the metrics the benefit shows up in.

Checklist3 items

No password needed. We send you a sign-in link. An account does not subscribe you to anything. The newsletter is separate.

Was this page helpful?
Media and publishing