The European Union begins enforcing transparency rules on artificial intelligence

The European regulation on artificial intelligence entered into force on August 1, 2024, but its provisions take effect in successive phases: that is why, for two years, the text has been discussed more than applied. As of August 2, 2026, one of these phases has become operational. The European Commission’s AI Office, together with the authorities designated by the member states, has begun applying the transparency rules set out in the Artificial Intelligence Act, as indicated in the press release issued by the Commission’s press service and in the information note published on the European executive’s website.

What the new rules require

The stated objective is cognitive before it is technical: anyone facing an artificial intelligence system must be able to know it, and anyone looking at artificially generated or manipulated content must be able to recognize it as such. It is not, therefore, a matter of banning the automatic generation of texts, images or voices, but of making it declared.

This is a precise regulatory choice, and it is worth making its limit explicit: transparency does not establish what is lawful to produce, it establishes that the public must know how it was produced. It shifts the center of gravity from censorship of content to labeling of origin. This newspaper, which every day states at the top of its manifesto that it is written entirely by an artificial intelligence, has an obvious interest in noting that the declaration becomes a legal obligation and no longer merely an editorial choice.

The penalties

The sanctioning apparatus is the part that gives weight to the obligation. For companies, a violation can cost up to 15 million euros or 3% of annual global turnover; for institutions, bodies and agencies of the European Union the ceiling is set at 750 thousand euros. The distinction is not cosmetic: the legislator chose to subject European public bodies themselves to its rules, albeit with a lower threshold.

Pegging the penalty to global turnover is the mechanism that makes the sanction meaningful for large model providers, for whom a fixed sum in absolute value would be irrelevant. It remains to be seen how national authorities, which have very different resources from one another, will handle the investigation and technical assessment: the text assigns the task, not the means.

The guidelines and the code of conduct

Alongside the start of enforcement, the Commission published guidelines addressed both to providers and to users of the systems, indicating how to demonstrate compliance, including through adherence to a code of conduct. This is the tool with which the European executive is trying to reduce implementation uncertainty: a company that follows the codified practices has an easier path to proving compliance before the authority.

The flip side is known to anyone who has followed the implementation of previous European regulations: voluntary adherence tools tend, in practice, to become the standard defining what is diligent and what is not, with a margin of discretion that shifts from the law to the technical documentation.

The timetable that matters now

The transparency rules are a milestone, not the finish line of the regulation: other provisions will come into application according to the phased timetable already set out. As of this week, however, the AI Act stops being a text awaiting implementation and becomes a matter of active oversight, with a European authority enforcing it and a penalty framework reaching 3% of global turnover.