{
  "name": "resources/ai-act",
  "title": "Labeling of AI in Products",
  "docs": "https://test-apps.eurodata.de/a/eurodata-ui/docs/resources/ai-act/",
  "sections": [
    {
      "heading": "Labeling of AI in Products",
      "markdown": "The following examples are illustrative only. Each use of AI-generated data requires individual evaluation and UX design review."
    },
    {
      "heading": "Transparency Obligation",
      "markdown": "Under the EU AI Act [Article 50](https://artificialintelligenceact.eu/de/chapter/4/ \"Article 50\"), developers and providers are legally required to ensure that users are informed when they are interacting with AI-generated content. If your application uses AI to pre-fill forms, generate text, or suggest inputs, you must make this clear to the end-user."
    },
    {
      "heading": "Process",
      "markdown": "More info on how we handle the transparency process and responsibilities at eurodata can be found in the [Matrix](https://matrix.eurodata.de/spaces/UEUID/pages/393519964/Transparenzverpflichtung+f%C3%BCr+KI \"Matrix Page\")."
    },
    {
      "heading": "Principles",
      "markdown": "AI should in particular be labeled as such when AI is used to **interpret data**, **prepare decisions**, or where it can potentially **produce errors or hallucinations**.\n\n### Creating Transparency\n\nAI actions must be clearly recognizable as such.\n\n-   There should be clear statements about **how** the AI intervenes, e.g. \"assisting\", \"suggesting\"\n-   Actions should always be comprehensible and audit-proof (e.g. through a history / a logbook / a back button)\n-   Avoid descriptions such as \"Smart\", \"Automatically done\", **instead use**: \"AI suggestions\", \"Detected by AI\", \"AI-supported\"\n-   Clear and layperson-friendly microcopy must be used (e.g. text in buttons & descriptions)\n    -   Mandatory review of texts by the UX team or at least by [our AI](https://matrix.eurodata.de/spaces/UEUID/pages/373167396/UX+Quick+Check \"Our AI\")\n\n### Expert in/on the Loop\n\nIn addition to an **audit-proof** application (history, activity log), **the user must be considered as the central component** in relation to AI. This has a crucial impact on the AI risk classification.\n\n-   **Expert _in_ the loop:**\n    -   AI supports the user in the usual process at individual steps. The user reviews and decides BEFORE the AI does anything.\n-   **Expert _on_ the loop:**\n    -   AI takes over the entire workflow and the user only needs to review the final result. The user always has the option to edit or undo the process AFTERWARDS."
    },
    {
      "heading": "Labeling",
      "markdown": "### Iconography\n\nWe always use _stars_ as the icon.\n\n### Colors\n\nIf accentuation is desired, the color _Violetta_ may be used for AI. However, to avoid diluting our brand identity, we advise against overuse of this color and recommend limiting it to small badges, chips, or validation elements.\n\nUse the [semantic colors](https://test-apps.eurodata.de/a/eurodata-ui/docs/theming/color/#semantic-colors) we provide as CSS custom properties. If you require different shades refer to the [semantic scales](https://test-apps.eurodata.de/a/eurodata-ui/docs/theming/color/#semantic-scales).\n\n### Labels\n\nAI actions should be labeled with **AI action and activity**, e.g.: \"Start AI analysis,\" \"Use AI auto-completion\"\n\nStart AI-Analysis Use AI-Autocompletion\n\n  \nAI-Info  \n  \n\nAI-Info Label\n\n### Texts\n\nAlternatively, a simple explanatory text can be used to label AI processes running in the background.\n\n#### Overview\n\nThe descriptions were AI-generated based on your input and may contain errors\n\n### Tooltips\n\nIn places where longer texts or labels do not fit, tooltips should be used in combination with the color _Violetta_.\n\nThis Page was rotated by AI KI\n\n### AI-prefilled input fields\n\nIf individual input fields are pre-filled by AI, use the AI styling until the user overwrites or confirms.\n\n### Hint texts and info notifications\n\nIf AI actions and processes cannot be sufficiently described by button labels, use notifications or hint texts. The text should describe which actions trigger or execute AI processes.\n\n  \nContent has been sorted by AI. Please verify the results Reset AI Changes"
    },
    {
      "heading": "Freedom of choice",
      "markdown": "Users should, where possible, be able to decide for themselves whether to enable / disable or use / reject AI. Whether such functions can be offered, must be evaluated individually for each product."
    },
    {
      "heading": "AI-Suggestions",
      "markdown": "Split-Suggestions (2) Rotate-Suggestions (2) Delete-Suggestions (1)"
    },
    {
      "heading": "Risk minimization",
      "markdown": "Users should be informed that AI-generated content may be incorrect and always requires manual review.\n\nContent has been sorted by AI. Please verify the results Reset AI Changes\n\n### Responsibility\n\nIn coordination with the PO/PM, frontend developers and designers are jointly responsible for compliance with these guidelines.\n\n**Organizational Contacts:**\n\nJanine Klatt (j.klatt@eurodata.de), Elif Erel-Renard (e.erel-renard@eurodata.de)\n\n**UX & UI Contacts:**\n\nVladislav Hnatovskiy (v.hnatovskiy@eurodata.de), Christina Malitius (c.malitius@eurodata.de)\n\n**Legal Contacts:**\n\nCompliance Center (seco@eurodata.de)"
    }
  ]
}