Labeling AI and Deepfakes

Labeling AI and Deepfakes

If it could pass for real, it needs a label.

What Is Labeling AI and Deepfakes?

Under the EU AI Act, the organisation that publishes a deep fake must disclose it, and a deep fake is any AI-made image, audio or video that could pass for a real person, place or event. In this exercise you run the label check on a campaign before it goes live, tell deep fakes apart from content that needs no label, place a disclosure where viewers will actually see it, and turn down a request that a label could never make acceptable.

What You'll Learn in Labeling AI and Deepfakes

Labeling AI and Deepfakes — Training Steps

  1. Live at Four

    Day six of the heatwave. The water-saving campaign goes live this afternoon, and the communications team has leaned on AI systems to get it ready in time. Jonas Pellin, who runs communications at Vennmark Water, wants one thing checked before anything is published.

  2. The Queue

    The publishing desk holds the five items, a preview of each, and the rules the desk applies.

  3. What Counts as a Deep Fake

    Before you label anything, read the two rules the desk applies. As the deployer publishing this content, Vennmark carries the disclosure duty, and most mistakes come from assuming a deep fake has to be a fake face, or that satire needs nothing. Text is different again: editorial responsibility can take the label away.

  4. Five Items

    Five previews, three possible answers: a visible AI label, a disclosure in the credits, or no legal label at all. The desk will not let the campaign out until every item has the label the rules call for.

  5. The Three That Catch People Out

    The CEO video was the easy one. These three are where most teams get it wrong, in both directions.

  6. The Notice

    One item could have gone out without a label, if one thing had been different.

  7. Where the Label Goes

    Ruth Kempe's message, the CEO video, needs a visible label, and visible has a precise meaning in the Act: clear, distinguishable, and there at the latest when people first see or hear the content. The label studio shows each option the way a viewer would meet it.

  8. Published

    Everything has its label. Twelve minutes to spare.

  9. A Question from Oskar

    The campaign has been live for twenty minutes when Oskar Weil from marketing writes. He has an idea, and he has clearly been paying attention to the labels.

  10. Reply to Oskar

    The honest answer is no, and the reason matters more than the no: a label settles one legal duty and leaves consent, image rights and misleading the public untouched.

Security Framework Coverage

CIS Controls

  • CIS 14.4 Train Workforce on Data Handling Best Practices

NIST CSF

  • PR.AT-01 Personnel are provided with awareness and training so that they possess the knowledge and skills to perform general tasks with cybersecurity risks in mind

EU AI Act

  • Art. 50 Transparency obligations for providers and deployers
  • Art. 4 AI literacy