Real numbers from this tool, line by line.
Up to €240,000 is the fine cap for an unlabelled AI chat — SaaS product designers at an EU SME with €8,000,000 turnover. This is the free checker's result for a SaaS with an AI chat assistant and AI image output, on the market before August, checked on 2026-09-27:
Fine cap (Art. 99(4) and 99(6)): up to €240,000 — 3% of €8,000,000 is lower than €15,000,000 Art. 50(1) tell people they are talking to an AI | from 2026-08-02 | applies — 56 days in force | AI chatbot notice (free), Assistant avatar, Human handoff button Art. 50(2) mark AI output machine-readable (system already on the market) | from 2026-12-02 | applies in 66 days | visible: AI content badge, Output label chip; metadata marking: not in a UI kit
Since 2 August 2026, Article 50 of the EU AI Act has applied to every SaaS product that talks to its users through an AI system. The rule in paragraph 1 is short: people must be told they are interacting with an AI system, at the latest at the first interaction. For a chat assistant that means before or at the very start of the conversation, not in the terms of service.
Article 99(4) sets fines for breaching Article 50 at up to €15,000,000 or 3% of total worldwide annual turnover for the preceding year, whichever is higher. Article 99(6) turns that around for SMEs and start-ups: whichever is lower. That is why the SME above lands at up to €240,000. A company that is not an SME with €40,000,000 turnover lands at up to €15,000,000, because the fixed amount is higher than 3% (€1,200,000).
The dates are not all the same. The machine-readable marking of AI-generated images, audio, video and text in Article 50(2) has one grace period, stated by the European Commission: systems already on the market before 2 August 2026 must comply from 2 December 2026. New systems have no grace period. The checker switches between the two dates when you tick whether your AI system was on the market before August.
A UI kit cannot do everything here, and we say so in the product. Machine-readable marking means metadata or a watermark inside the file; that is engineering work, not a component. What a design file can do is the visible part: the notice at the start of a chat, the label next to an AI-generated image, the disclosure on a deep fake or on AI text published to inform the public, and the button that gets a user to a person.
That is what the SaaS UI kit contains, counted: 8 components and 26 variants. The AI chatbot notice has 4 variants — chat start in English, chat start in German, a voice agent notice and the handoff to a human. The AI content badge has 4 (AI-generated, KI-generiert, Edited with AI, Human-reviewed). The output label chip has 3, the assistant avatar 3, the AI settings toggle 3, the "Why am I seeing this" tooltip 3, the human handoff button 3 and the AI answer card 3.
It ships three ways. A Figma plugin (manifest.json plus code.ts) builds the 8 components as real variant sets with text; you compile code.ts with tsc and import it as a development plugin. 26 SVG files paste into Figma as vectors. One .sketch file opens in Figma by drag and drop and in Sketch directly; it holds the variants as named, coloured shapes without text layers, so the labels come in through the SVGs or the plugin. Swap the colour tokens for your brand colours and rename the assistant "Ava". The checker is not legal advice; it applies the fine formula in Articles 99(4) and 99(6) and the dates in Article 113 and the Commission FAQ.
The free part is complete for one job: the four chatbot notice variants plus the checker. If your product only has a chat assistant, that is the Article 50(1) screen, done. The full kit is for what the notice leaves open once the AI also writes and draws for your users: the labels on its output, the avatar that shows who is answering, the toggle that turns AI suggestions off, and the path to a person. It is one payment, not a subscription.