The European Union began enforcing the AI Act’s transparency rules on Sunday. Chatbots operating in the bloc must now tell users they are talking to a machine, and AI-generated deepfakes require clear labels.
The European Commission’s AI Office and national regulators also gained enforcement powers for the first time. Penalties reach €35 million or 7% of global annual turnover for the most serious violations.
Note: A normal person posting an AI-generated image on a personal social media account would not be fined under this EU rule. Personal, non-professional use is excluded from the AI Act. The situation changes when the content is used professionally or commercially. For example, by a business, freelancer, or monetised influencer.
What the EU AI Act Now Requires
The Commission confirmed that Article 50, the law’s transparency chapter, applies from August 2, 2026. AI systems that interact directly with people must reveal they are machines. The duty covers chatbots, voice assistants, and agents from the first interaction onward.
The rules apply to any provider or deployer whose system reaches users in the EU, regardless of where the company is based.
The duty extends beyond conversation. Deployers must flag AI-generated or manipulated images, audio, and video as artificial. Text published to inform the public also needs a label unless a human editor has reviewed it and taken responsibility.
Companies running emotion recognition or biometric categorization systems must inform every person exposed to them. An independent guide to the provision notes that clearly creative or satirical uses face lighter disclosure duties.
One element got extra time. Generative systems already on the market have until December 2, 2026, to add machine-readable watermarks to synthetic content.
Regulators Can Finally Issue Fines
Until now, the AI Act operated largely on trust. General-purpose AI model providers have carried documentation and copyright obligations since August 2025. However, Brussels had no power to compel compliance.
That changed on Sunday. The AI Office may now demand documentation, evaluate models directly, order corrective measures, or pull models from the EU market. Transparency breaches carry fines of up to €15 million or 3% of worldwide turnover.
The stakes rise for prohibited practices, where penalties climb to €35 million or 7%. The shift lands as Europe’s play for Anthropic shows the bloc courting the same firms it now polices.
Most of the Feared Deadline Never Arrived
August 2 was long billed as the EU AI Act’s biggest compliance date. The Digital Omnibus, an amendment package signed July 8, postponed the high-risk obligations due the same day.
Hiring, credit scoring, and law enforcement systems now have until December 2027. AI embedded in regulated products, such as medical devices, has until August 2028.
Lawmakers framed the delay as time for technical standards to mature, while critics called it a retreat under industry pressure. Developers have pushed back on rules globally, recently backing open AI models against proposed limits.
| Change | What it means | Effective |
|---|---|---|
| Chatbot disclosure | AI systems must identify themselves to users | August 2, 2026 |
| Deepfake labels | AI-generated media must be disclosed as artificial | August 2, 2026 |
| Enforcement powers | Fines up to €35 million or 7% of turnover | August 2, 2026 |
| Content watermarking | Machine-readable marks on synthetic content | December 2, 2026 |
| High-risk systems | Hiring, credit, and policing AI obligations | December 2, 2027 |
| High-risk products | AI in medical devices and machinery | August 2, 2028 |
The surviving rules may matter most for crypto. AI trading bots, automated support agents, and token projects using AI-generated promotional videos all fall under the disclosure duties.
The first enforcement actions will show how hard the AI Office intends to swing.
Bitrue