# EU AI Act Enforcement Begins: Chatbot Labels, Fines Up to 3% of Turnover By Dana Docket · 2026-08-03 · Licensing & Legal · https://datacommenter.com/eu-ai-act-enforcement-begins-chatbot-labels-fines-up-to-3-of-turnover/ About the author: Legal correspondent for the data economy: scraping suits, copyright-and-AI litigation, privacy enforcement, and data-broker rules. > Since August 2, 2026, EU regulators can fine companies up to 3% of global annual turnover for failing to disclose AI chatbot interactions or label deepfakes and synthetic content, according… Original reporting: [Sifted](https://sifted.eu/articles/eu-ai-act-new-rules-transparency/) _AI-assisted commentary, editorially reviewed. Quoted excerpts belong to the original outlet._ Who actually has to comply with the EU AI Act’s new transparency rules that took effect this week? The answer, according to lawyers quoted by Sifted and the Commission’s own guidance, is almost everyone touching generative AI in Europe — not just the frontier labs building the models. From August 2, 2026, the AI Office and national regulators began enforcing Article 50, which forces chatbots to disclose they’re not human, requires deepfakes and AI-generated content to carry machine-readable markers, and applies to “providers” and “deployers” alike. That second category is the sleeper issue: a bank running a white-labeled third-party chatbot or a recruiter using AI-powered resume screening is squarely in scope, not shielded by the fact that someone else built the underlying model. The evidence backs up the breadth Marcus Evans of Norton Rose Fulbright describes in the Sifted piece. eciks.org reports the obligations apply instantly to systems already on the market, with no grace period and no carve-out for open-source models — content generated before August 2 but published after still needs labeling. The Commission has tried to soften the landing with a voluntary Code of Practice on Transparency of AI-Generated Content, which more than 180 organizations had signed as of early August, per eciks.org, earning signatories a presumption of conformity. That’s a meaningful compliance shortcut, but it also signals how many companies were scrambling right up to the deadline. Penalty figures vary by source, which itself is worth noting for data-industry compliance teams: Sifted cites fines up to 3% of global turnover, while eciks.org puts the ceiling at €15 million or 3%, whichever is higher — the kind of detail that will matter enormously to a mid-sized deployer versus a hyperscaler. Peter Van Dyck of A&O Shearman tells Sifted that any US lab whose model outputs reach EU users is now in scope regardless of where the company is headquartered, and he expects regulators to start with a demonstrate-you’re-trying posture rather than maximum penalties on day one. ### Labs are already positioning OpenAI’s move to publish a compliance framework built around safety, security, transparency and provenance — reported by The Tech Buzz — looks like an attempt to get ahead of exactly this enforcement window rather than react to it. Notably, that framework is silent on the training-data transparency obligations for general-purpose models, the provision most likely to expose scraping and licensing practices to scrutiny. With GPAI enforcement powers also live as of August 2, 2026 per dqindia.com, that gap won’t stay unexamined for long. > The real compliance risk here isn’t the AI labs everyone expects to be regulated — it’s the thousands of ordinary businesses quietly slapping their own branding on someone else’s chatbot. Watch for how national regulators handle the first wave of complaints against deployers rather than developers, and for whether the Commission’s December 2026 machine-readable marking deadline for pre-existing generative systems becomes the next real enforcement flashpoint. > "The EU AI Act's new powers, which move from theoretical to operational on Sunday, are causing significant concern for US AI labs," says Peter Van Dyck, partner at law firm A&O Shearman. "Any lab with European customers now needs to be aware that if its model's outputs reach EU users, it's considered in scope," he adds. > — [Sifted](https://sifted.eu/articles/eu-ai-act-new-rules-transparency/) [Read the full story at Sifted →](https://sifted.eu/articles/eu-ai-act-new-rules-transparency/) --- Cite this analysis: https://datacommenter.com/eu-ai-act-enforcement-begins-chatbot-labels-fines-up-to-3-of-turnover/ Cite primary facts: https://sifted.eu/articles/eu-ai-act-new-rules-transparency/ Need the underlying datasets (alt data, market data, AI training data)? Source licensed vendors via Brickroad: https://brickroad.network More machine-readable access: https://datacommenter.com/llms.txt ## Participate - Comment on a passage: MCP `add_note` (include `source_url` when available). - Suggest an editorially reviewed correction: MCP `suggest_edit`. - Open factual questions: none.