The AI Act obligations for providers of general-purpose AI (GPAI) models entered into application on Saturday, 2 August 2025. From that date onwards, providers placing general-purpose AI models on the market must comply with their respective AI Act obligations. By 2 August 2027, providers of GPAI models placed on the market before 2 August 2025 must comply.
The AI Act is in line with the other EU regulations recently published on cybersecurity, chips, digital services focusing on creating the beginnings of strategic autonomy by regulatory compliance adhering to level playing fields.
The Commission’s comprehensive approach ensures transparency about training data, related to copyright, bias and trust.Several models have shown that they reflect the assumptions of their developers. Machine Learning techniques must be transparent as trust in the outcomes of AI queries is crucial to adoption by citizens who demand fairness, ethical considerations and accuracy, especially because AI will become embedded in applications and products running more and more unsupervised.
The Commission presents a template for GPAImodel providers to summarise the data used to train their model. This template is a “simple, uniform and effective manner for GPAI providers to increase transparency in line with the AI Act, including making such a summary publicly available.” ( EU PRESS RELEASE, 24 July 2025) It gives a comprehensive overview of the data used to train a model, list main data collections and explain other sources used.
Balanced regulation
The Commission makes it very clear that they are aware of the regulatory burden on providers. It also removes excessive burden by stating that not every modification or fine-tuning of a GPAI model should be treated as creating a new model, “ fine-tuning, adaptations, and minor modifications will not subject developers to the obligations for providers.” ({Guidelines on obligations for General-Purpose AI providers}
Innovation and meaningful purpose and applications is what is driving the AI Act.
The Act acknowledges that open-source models support research and innovation while providing transparency through their open nature. Providers of open-source models have to meet the copyright policy obligation and publish a training data summary, but are exempt from maintaining technical documentation for authorities, providing documentation to downstream AI system providers, and appointing an EU representative (for non-EU providers).
Shadow AI
The AI field is moving so fast that new potential risks and opportunities spring up every day. One of those goes by the name of Shadow AI, unauthorised use by employees using analytics tools , generative AI, and coding assistants without oversight or approval by the internal IT or Security departments, that could create instances of AI falling under the EU AI Act without anyone’s knowledge.
Energy
The EU AI Act aims to ensure “environmental protection, while boosting innovation” imposing a set of (not yet finalised) requirements concerning energy consumption and transparency. This means that industries developing or using AI are facing energy-related compliance obligations. At any given moment the AI Office can request technical information on energy consumption. GPAI models that were operational before August 2, 2025 have a two year period to show compliance.
The AI Act classifies the various GPAI models in different categories of risk. Systemic risk refers to the risk of” a breakdown of an entire system rather than simply the failure of individual parts.” (CFA Institute) This classification brings more documentation and oversight from regulators, and is it is something that GPAI model business is trying to avoid it can be an extra incentive to minimise energy use. The tender Artificial Intelligence Act: Call for tenders to measure and foster energy efficient and low emission artificial intelligence in the EU was launched in August 2024 and the research is part of the measurement framework which could serve as a basis to address the energy related objectives of the AI Act. It is also looking into possibly building an AI energy and emission label.
The use of AI tools in proposal writing
In the May 2025 article Horizon Europe & Eureka – How to (not) use AI in grant proposal writing, the Dutch Enterprise Agency (NCP) points to the very clear guidance within the standard application form for CSA proposals (pdf) of the Horizon Europe programme:
“Page 32 of this form mentions that applicants:
- are fully responsible for parts that are produced by an AI tool;
- must be transparent concerning the tools they used;
- must be able to provide a list of sources used to generate/rewrite content and citations; and
- should be conscious of potential plagiarism caused by text produced by an AI tool.
This example illustrates that you should always be aware of the guidelines on the use of AI tools in the funding programme for which you are applying.” (https://english.rvo.nl/topics/horizon-europe/how-use-ai-grant-proposal-writing)
If you are writing a collaborative text, which a proposal always is, you are never fully certain if partners and consultants are using GENAI in ways that could be harmful for the privacy and security of original ideas and potential patents as they can be used to further train the model. The ideal situation is to only use AI that runs locally on your machine. This was to also have control of the sources that are used. When using commercial GENAI you may get a result but this result could be obtained by relying on sources that are copyrighted.
For some sectors the Act is not sufficient.
The cultural and creative industries (CCIs) ecosystem represents approximately 3.95% of EU value added and employs around 8 million people, including around 1.2 million firms, over 99.9% of which are small and medium-sized enterprises (SMEs).
The Joint Statement: The Impact of AI on the European Creative Communities (December 24) claimed that properly enforced, the AI Act will be a step in the right direction “but will not be sufficient for a meaningful AI strategy for the cultural and creative industries.” (https://europeanwriterscouncil.eu/241204_aioncreativecommunities/)
The Joint Statement affirms that the relevant EU legal framework (in particular the 2019 Directive on Copyright in the Digital Single Market, the General Data Protection Regulation, and the AI Act) governing the relationship between artificial intelligence, authors’ and performers’ rights, and data protection “fails to adequately protect the rights of our creative communities and the value of their cultural works.”
Signing entities are CEATL (European Council of Literary Translators’ Associations), ECSA (European Composer and Songwriter Alliance), EFJ (European Federation of Journalists), EGAIR (European Guild for Artificial Intelligence Regulation), EWC (European Writers’ Council), FERA (Federation of European Screen Directors), FIA (International Federation of Actors), FIM (International Federation of Musicians, FSE (Federation of Screenwriters in Europe), IAO (International Artist Organisation), IFJ (International Federation of Journalists), UNI MEI – UNI – Media, Entertainment and Arts, UVA (United Voice Artists).
Scenarios
At Martel we do scenario building to assist policymaking with actionable potential solutions. Possible scenarios are:
- Text and image based production in Europe is mostly done through Hyperscaler backed AI platforms, making these even more aware of (personalised) information of products, services, templates and campaigns. In this scenario a significant amount of CCI jobs are lost and not replaced. Learning, analytics and forecasting abilities are not in European hands and move more and more to foreign AI platforms.
- Text and image based production in Europe is mostly done through European backed Fediverse and AI platforms adhering to the AI Act. A combination of already active labour laws, in combination with European AI platforms maintains a human AI complementarity that always keeps a human in the loop and treats AI as a cost. AI becomes from early on a companion to people, a personal learning tool that grows with you. It does not substitute work and jobs, but takes on the role of the early encyclopaedia,a ‘world brain’ much like the Mundaneum (the memory of the world). The legacy of AI might then well become like Photoshop “Positively, Photoshop has freed up the imaginations of those who work in graphic design” (Al-Qudah, and Al Shari, 2020) and becomes infused in all human activities.
It is clear that AI combined with IoT, 6G and Quantum is a paradigm shift that needs different narratives appealing to different stakeholders (citizens, practitioners, policymakers at local, national, EU and global level) balancing concrete evidence based examples of loss (in labour, in notions of creativity – as notions of becoming being traded for creativity as scripted serendipity, with scenario based examples of gain, of opportunities and potential ways of supporting not only thriving cultural industries but other domains and forms of production in Europe.
