As of early August, new transparency rules for AI-generated content are in effect, while the EU simultaneously launches three major projects worth €36.8 million to develop its own GenAI solutions in public administration.
The European Union has made a double leap forward in the field of artificial intelligence: on one hand, it tightens the rules of the game for tech giants, and on the other, it invests tens of millions of euros in developing its own sovereign AI tools. The key question is whether European ambitions in generative artificial intelligence (GenAI) can survive contact with a market dominated by American and Chinese players.
As of August 2, 2026, new transparency rules stemming from the Artificial Intelligence Act (AI Act) are in force across the European Union. Their goal is to help citizens clearly recognize when they are interacting with an AI system or are exposed to content generated by one.
AI service providers must now clearly and visibly label content created or manipulated by artificial intelligence, including machine-readable labels. This applies to images, audio, and video that imitate real people, objects, places, or events, as well as texts on matters of public interest that have not undergone human editorial control.
Non-compliance carries hefty fines. Companies can be fined up to €15 million or 3% of global annual turnover, while EU institutions and agencies can face fines of up to €750,000. Enforcement will be overseen by national regulatory bodies, the European AI Office, and the European Data Protection Supervisor.
For journalists and newsrooms, the key information is that not every use of AI tools triggers an automatic labeling obligation. According to the interpretation of the Independent Journalists' Association of Serbia (NUNS), the decisive factor is whether the content has undergone substantive human review and whether someone has taken editorial responsibility for its publication. Text generated or modified by AI and published to inform the public does not need to be labeled if it has passed appropriate human control.
However, the European Commission warns that a superficial check of spelling or grammar is not sufficient. An editor must have the authority to approve, modify, or reject content based on its substance. On the other hand, deepfake content must be labeled even when it has passed editorial control.
To facilitate implementation, the Commission has also published a Code of Practice on transparency of AI-generated content. Although voluntary, the European Artificial Intelligence Board has deemed it an appropriate framework for demonstrating compliance with the law.
In parallel with regulatory tightening, Brussels has announced three new projects to boost generative artificial intelligence in public administrations. The total value of the projects is €36.8 million, of which the EU directly provides €18.4 million.
The first project, "Floods and droughts," is worth €8.8 million (with €4.4 million from the EU) and is led by Italy in collaboration with France, Spain, and Greece. It is expected to be completed by June 30, 2030.
The second project, "Eunomia AI," has a total value of €13.9 million (with €6.9 million from the EU) and is led by Spain. The consortium includes 13 countries of the European Economic Area, including Greece, Austria, Estonia, Finland, Portugal, Latvia, Germany, the Netherlands, Poland, Hungary, Norway, and Malta. The goal is to implement GenAI in 25 public administrations and 15 public services within a year and a half. The expected completion date is June 30, 2029.
The third project, "EuropAI," also worth €13.9 million (with €6.9 million from the EU), is led by the Netherlands and also includes Denmark, Belgium, and Luxembourg. The core of the project is the creation of a centralized GenAI infrastructure at the University of Granada, with completion expected by June 30, 2029.
The EuropAI project team, as reported by author Egle Markeviciute for Euronews, vividly describes the current situation: European public administrations currently rely on imported "AI fast food" from non-EU providers like ChatGPT, Copilot, or Gemini, which they describe as expensive, non-transparent solutions misaligned with European sovereignty and compliance requirements.
In contrast, the EuropAI team sees its project as a "shared kitchen where member states jointly create, test, and share GenAI solutions." The goal is to create "ready-to-serve meals" with clear "nutritional labels" in an "organized pantry." The Danish platform OS2ai, which already serves 28,000 employees, and the Dutch GovChat-NL are examples of existing national solutions that could be replicated at the European level.
Although Croatia is not explicitly listed as a member of the consortium in these projects, all obligations from the AI Act directly apply to it as an EU member. The uniform transparency rules create, as Brussels puts it, "a single market and harmonized rules for trustworthy AI in the EU."
For autumn 2026, the first EU summit dedicated exclusively to artificial intelligence is planned, which will be an opportunity to define further steps in the European AI strategy.
