Skip to main content
Ga naar de homepage van Developers

AI Transparency & Disclosures

What is the standard for Transparency & Disclosures?

This standard ensures that end-users are fully aware when they are interacting with an AI system or consuming AI-generated content. It fulfills the transparency obligations of the AI Act by requiring clear communication, visible markings, and comprehensible explanations of the AI's role and limitations.

When and for whom is this standard applicable?

This standard applies to UX/UI Designers, Frontend Developers, AI Engineers, and Product Owners involved in the design and deployment of user-facing AI applications within the municipality of Amsterdam.

What is required?

To ensure trust and legal compliance, the following transparency measures must be implemented:

1. Interaction Disclosure

  • Clear Notification: Users must be explicitly informed that they are interacting with an AI system rather than a human. This applies to chatbots, automated decision systems, and conversational agents.
  • Persistent UI Elements: Use persistent UI banners, introductory disclaimers, or watermarks to maintain awareness throughout the user's interaction with the system.

2. AI-Generated Content Marking

  • Human-Readable Labels: Any content (text, audio, image, or video) that is generated or significantly manipulated by an AI system must be clearly labeled as such (e.g., "This report was generated by AI").
  • Machine-Readable Metadata: Where applicable (especially for media and deepfakes), embed machine-readable metadata or cryptographic watermarks to establish the provenance of the AI-generated content.

3. Explainability and Plain-Language Communication

  • System Logic Explanation: Provide users with access to a plain-language, non-technical explanation of how the AI system works, what data it uses, and the general logic behind its recommendations or decisions.
  • Limitation Disclosure: Clearly communicate the known limitations of the system to the end-user (e.g., "This system's data is accurate up to 2024 and may produce incorrect information").

What to avoid?

  • Designing interfaces that anthropomorphize the AI to deceive users into believing they are speaking with a real human.
  • Hiding AI disclosures in deep, hard-to-reach Terms of Service pages instead of the primary user interface.
  • Passing off AI-generated reports or decisions as solely human-authored without acknowledging the AI's contribution.

Considerations

  • Transparency measures should be designed in conjunction with the Human Oversight & Operational Boundaries to ensure users know how to escalate issues to a human if they disagree with the AI's output.
  • Align UI disclosures with the Amsterdam Design System and general Frontend Accessibility standards to guarantee notifications are readable by screen readers and visible to all users.