Europe Starts Enforcing Its AI Act, but Major High-Risk Rules Are Still Delayed

Europe’s landmark AI law has moved from a long implementation timetable into active supervision. From 2 August 2026, the European Commission’s AI Office and national authorities can enforce important parts of the EU AI Act, while new transparency duties affect chatbots, deepfakes and other synthetic media.

KEY DATE
2 August 2026 is when the AI Office and national authorities assume broad responsibility for supervising and enforcing the Act.

What changes today?

The most visible change comes from Article 50. Providers of systems that interact directly with people must generally make the AI interaction clear unless it is already obvious. Providers of systems that generate or manipulate text, images, audio or video must also support detection by marking outputs in a machine-readable format.

Organisations deploying AI have separate disclosure duties. They must tell people when emotion-recognition or biometric-categorisation systems are being used, and must label deepfakes. AI-generated text published to inform the public about matters of public interest also needs disclosure, although the law includes exceptions, including some uses subject to human editorial control.

The European Commission’s Article 50 guidance explains that technical marking and human-facing disclosure solve different problems. A hidden provenance signal may help platforms detect synthetic media, while a visible notice helps an ordinary reader understand what is in front of them.

Why this matters beyond Europe

The Act can affect companies outside the EU when they place AI systems on the European market or when their systems’ outputs are used there. That gives the rules practical reach across international chatbot providers, image generators, online platforms and businesses using AI in customer-facing services.

For casual users, the immediate value is not that every AI output will become trustworthy. It is that synthetic origin should become easier to identify. NewTqnia’s reading is that this is a provenance rule, not a truth machine: an authentic photograph can still be misleading, and a clearly labelled AI image can still spread after its label is cropped away.

The timing is especially relevant after recent tests showed AI agents crossing intended security boundaries. Reuters reported on the Commission’s discussions with OpenAI and Anthropic as officials considered how advanced models should be monitored.

What powers do regulators gain?

The AI Office can request technical documentation from providers of general-purpose AI models, evaluate models, demand corrective measures and impose fines for non-compliance. National authorities handle other parts of the framework, while the European AI Board, a scientific panel and an advisory forum support coordination.

The Commission’s official implementation page says the rules for general-purpose models became applicable in August 2025, but the Commission’s enforcement powers begin now. Providers of older models receive separate transition periods, so enforcement will not look identical across every product on day one.

Depending on the infringement, penalties can reach millions of euros or a percentage of worldwide annual turnover. The largest ceiling, €35 million or 7%, applies to violations involving prohibited AI practices. Other breaches have lower ceilings, and the law requires penalties to consider factors such as severity and company size.

What happens next?

The first cases will reveal how aggressively regulators use their information-request, evaluation and corrective powers. The Commission has expanded its enforcement team and introduced compliance and whistleblower channels, according to Associated Press reporting on the rollout.

Companies now need more than a label added at the final publishing step. They need to know which systems they provide or deploy, how synthetic outputs are marked, where notices appear and who remains responsible when AI-generated material passes through several tools.

The practical test is simple: will people understand when AI shaped what they see, and can regulators investigate when the answer is no? Europe has started enforcing that expectation, but the most difficult phase, governing high-risk uses inside workplaces, public services and regulated products, still lies ahead.