Articles

The provisions of the EU AI Act coming into effect in August may not change journalism overnight, but they will have a significant impact. The most consequential change relates to transparency: now it is a legal obligation.

On August 2, the EU AI Act reached a new milestone. The provisions entering into force and becoming fully enforceable in practice do not focus primarily on regulating AI use in journalism. Instead, they concentrate on disclosure, labelling, and enforcement: they introduce a regulatory layer into editorial decision-making, particularly in areas where the boundaries between human and machine input are difficult to draw.

Disclosure, Exceptions, and Editorial Responsibility

As Krisztina Rozgonyi, senior scientist at the Austrian Institute of Technology and the Austrian Academy of Sciences, explains, Article 50 includes important elements. The most important provision for journalism is the disclosure obligation: if texts published to inform audiences on matters of public interest are generated with AI, they must be labelled accordingly, with certain icons. There are, however, certain exceptions: the disclosure obligation becomes more flexible if the content serves artistic or satirical purposes, and there is no labelling requirement if the text undergoes substantial human review and is subject to editorial responsibility.

This means that routine applications, such as automated summaries, translation, or the use of generative tools in drafting, may fall within the scope of disclosure requirements, depending on the degree of human editorial control.

This threshold, however, remains open to interpretation: it is vague what constitutes ‘meaningful editorial control’. According to Rozgonyi, regular, professional editorial practices fall into this category, and another EU regulation, the European Media Freedom Act, defines what editorial control means. Accountability procedures must be defined, responsibilities must be clear on the individual level. This could influence workflows, she continues, and maybe editorial policies should be revised. In fact, in the case of smaller outlets, this might be the time when they have to put these editorial policies into written form.

Furthermore, according to Article 50, deepfake content, which resembles real persons or events and may reasonably be mistaken, must be labelled accordingly. Users must be informed if they interact with AI agents, for example, chatbots. The European Commission’s guidance elaborates on these requirements and their intended scope.

As Rozgonyi argues, news organisations also need to be mindful whether they qualify as providers or deployers of AI systems. In most cases, they are deployers, however, if they build their own AI systems, for example news assistants or chatbots, then they have to align with further regulations as providers.

From Regulation to Enforcement

August 2 marks the date at which enforcement mechanisms must be in place across member states. Regulatory bodies, penalties, and oversight structures move from design to implementation. For media organisations, this turns questions of AI use into matters of compliance. Decisions about disclosure are no longer confined to internal editorial guidelines but carry potential legal consequences.

The broader ecosystem is also affected. Rules governing general-purpose AI systems have been in force since 2025, requiring providers to document their models and meet transparency standards. As enforcement intensifies, these obligations become more visible in practice. Journalists gain access to additional information about how such systems are trained and operate, which may inform reporting on issues such as sourcing, bias, or intellectual property.

At the same time, media organisations remain largely dependent on external technology providers, while bearing responsibility for how these tools are deployed in editorial contexts. The Act does not fundamentally change this relationship, but it does bring greater attention to it.

Still, some of the most far-reaching elements of the AI Act are not part of this phase. Provisions concerning high-risk AI systems, including those relevant to employment or certain forms of automated decision-making, are largely deferred to later implementation stages. For journalism, this means that the immediate regulatory impact is concentrated on content and transparency, rather than on the internal use of AI in business or organisational processes. According to Rozgonyi, it is yet to be seen what kind of impact these provisions may have later.

The result is an uneven regulatory landscape. Editorial uses of AI are subject to new expectations, while other applications within media organisations remain, for now, less directly affected. Furthermore, Rozgonyi believes that the bigger focus should be placed on improving AI literacy of audiences.

There are, however, emerging areas of opportunity. The Act requires member states to establish regulatory sandboxes, designed to allow organisations to test AI systems under supervision. For media organisations, particularly those with the capacity to engage in experimentation, this may provide a structured environment to develop and refine AI-driven practices in dialogue with regulators.

Why This Matters for Funders

For funders and other stakeholders in the journalism ecosystem, these questions intersect with ongoing concerns about sustainability, credibility, and the capacity of news organisations to adapt to technological change. As previous waves of transformation have shown, the introduction of new tools rarely affects only production processes but also reshapes relationships between journalists, audiences, and the institutions that support them.

Supporting journalism in an AI-shaped environment is no longer only about financing content or institutions, but about enabling the conditions under which trust can be maintained. That includes investment in editorial standards, disclosure practices, and technical literacy within newsrooms, as well as independent scrutiny of the tools those newsrooms rely on. Without that layer of support, the burden of compliance and transparency risks falling unevenly across the sector, deepening existing gaps between well-resourced organisations and those already struggling to stay afloat. In this sense, funding decisions begin to influence not just what journalism is produced, but how accountable, legible, and resilient it remains as AI becomes embedded in its core processes.

AI is already embedded in many aspects of journalism. What the Act does is bring greater structure to its use, particularly where that use intersects with public trust. In the coming years, the practical significance of these provisions will depend on how they are interpreted and applied. Much will be determined not only by regulators, but by the choices made within newsrooms and across the broader media ecosystem.