KI-Transparenz und Offenlegung

KI-Transparenz und Offenlegung

Wenn die Leute nicht erkennen können, dass es sich um KI handelt, müssen Sie es ihnen sagen.

Was ist KI-Transparenz und Offenlegung?

The EU AI Act sets transparency duties for chatbots, AI-generated content and deep fakes. In this exercise you review a customer-facing chatbot that hides its AI identity, then work through a content queue: you tell the legal deep fake disclosure apart from your company's own labeling policy, and settle a gray-area request about AI-polished text.

Was Sie lernen in KI-Transparenz und Offenlegung

KI-Transparenz und Offenlegung — Trainingsschritte

  1. Artikel 50: Transparenzpflichten

    Article 50 of the EU AI Act sets transparency duties that apply whatever an AI system's risk class: Chatbots must be designed so people know they are dealing with AI, unless it is obvious. That duty sits with the provider that builds them. Synthetic content generated by AI must be marked by the provider in a machine-readable way, so it can be detected as AI-made. Deep fakes must be disclosed by the deployer that publishes them: AI-made or altered images, audio or video that resemble real people, places, objects or events and could pass for authentic. Evident satire or art still discloses, in a way that does not spoil it. AI text published to inform the public must be disclosed too, unless a person reviewed it and someone holds editorial responsibility. These are legal requirements, not good practice, and breaching them can be fined.

  2. Prüfung des Kampagnen-Launches

    Eine E-Mail von Sarah Mitchell, der Marketing Director, trifft ein. Die Kampagne startet am Freitag, und mehrere Deliverables benötigen eine Compliance-Prüfung, bevor sie online gehen.

  3. Den Chatbot testen

    Alice folgt Sarahs Link zur Chatbot-Vorschau. Sie muss sich zunächst im CMS anmelden und sieht anschließend genau, wie eine Person den Chatbot beim ersten Besuch erlebt.

  4. Die Chatbot-Offenlegung korrigieren

    Der Chatbot präsentiert sich als Mensch ohne jeden KI-Hinweis. Alice muss die Konfigurationsseite des Chatbots im CMS öffnen und den verpflichtenden KI-Offenlegungs-Hinweis aktivieren.

  5. Content-Prüfungs-Queue

    With the chatbot disclosure fixed, Alice moves on to the content labeling queue. Five marketing assets need labels before they can be published. Cascade's publishing policy labels every asset by how much AI went into it, which goes further than the law. The one label the AI Act itself demands in this queue is the deep fake disclosure.

  6. Inhalte kennzeichnen

    Alice arbeitet jedes Content-Element ab und wählt jeweils die passende KI-Transparenz-Kennzeichnung anhand der Entstehungsweise. Der Fortschrittsbalken zeigt, wie viele Elemente bereits gekennzeichnet sind.

  7. Wissenscheck: Audio-Aufbereitung

    Before moving on, a question about where the law ends and company policy begins.

  8. Die Grauzone

    Eine E-Mail von Tom Becker, einem Kollegen, trifft ein. Er fragt, ob eine Pressemitteilung eine KI-Kennzeichnung benötigt. Den gesamten Inhalt hat er selbst geschrieben, den Entwurf jedoch durch ein KI-Tool laufen lassen, um „die Grammatik zu polieren und den Tonfall anzupassen“. Alice muss entscheiden, wo dies auf dem Spektrum der Offenlegung einzuordnen ist.

  9. Tom antworten

    Tom did the writing himself, but he let AI reshape sentence structures and word choices to change the tone. The AI Act asks for no label on text a person wrote and publishes under the company's editorial responsibility. Cascade's own policy does label AI help of this kind, so Alice draws that line for him clearly.

  10. Abschluss: Transparenz ist die Grundregel

    Alice has completed the compliance review and sent her recommendation to Tom. Here are the key takeaways from Article 50: Chatbot disclosure A chatbot must be built so people know they are talking to AI, unless that is obvious. The duty sits with the provider, and a human name and persona with no disclosure breaks it. What the Act requires, and what policy adds The Act asks providers to mark synthetic output machine-readably, and deployers to disclose deep fakes and AI text published to inform the public unless a person holds editorial responsibility. Cascade's policy labels more than that, by how much AI went into each asset. Deep fakes AI-made or altered media that resembles a real person, place or event must be disclosed by the organisation that publishes it. Evident satire or art still discloses, in a way that does not spoil it. When in doubt, disclose The cost of unnecessary disclosure is near zero - a small label or notice. The cost of missing a required disclosure is a regulatory violation. When the level of AI involvement is ambiguous, always err on the side of transparency.

Abdeckung der Sicherheits-Frameworks

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