The European Parliament's Internal Market and Civil Liberties committees adopted their joint position on a simplification proposal amending the Artificial Intelligence Act on Wednesday. The committees voted 101 in favor, 9 against, with 8 abstentions on the proposal. The proposal supports postponing the activation of certain rules on high-risk AI systems due to key standards potentially not being finalized by the current deadline of August 2, 2026. MEPs propose fixed application dates: December 2, 2027 for high-risk AI systems specifically listed in the regulation, including biometrics, critical infrastructure, education, employment, essential services, law enforcement, justice, or border management, and August 2, 2028 for AI systems covered by EU sectorial legislation on safety and market surveillance.
Specific provisions in the AI Act simplification include extended timelines for watermarking requirements and new restrictions on harmful applications. MEPs propose giving providers more time to comply with rules on watermarking AI-created content, suggesting an extension until November 2, 2026 instead of February 2, 2027 as proposed by the Commission. The committees also want to introduce a ban on 'nudifier' systems that use AI to create or manipulate sexually explicit or intimate images resembling an identifiable real person without consent. This ban would not apply to AI systems with effective safety measures preventing users from creating such images. Additionally, MEPs favor allowing service providers to process personal data to detect and correct biases in AI systems with safeguards to ensure it is done only when strictly necessary.
Support measures for businesses form another key component of the parliamentary proposal. MEPs backed extending support measures to small mid-cap enterprises to help EU companies scale up as they outgrow SME status. They also argue that obligations under the AI Act can be less stringent for products already regulated under sectoral laws, such as medical devices, radio equipment, or toy safety, to prevent overlapping application. The Commission would address regulatory gaps by updating those existing rules rather than creating duplicate requirements.
The first thing that strikes me is that it's not funny. And then I thought, 'Why does it say Øverli in the corner there?'
Parallel to the AI Act developments, the European Parliament adopted a series of recommendations on Tuesday urging lawmakers to find a permanent solution to protect copyright from use by AI. The report calls for EU copyright law to apply to all AI systems made available to users within the EU. It proposes creating a European register at the EUIPO listing every copyrighted work used to train AI models and artists who have opted out of having their work used. The report also suggests that companies disclose which websites they have scraped for training data.
Parliamentarians warn that failing to comply with transparency requirements could be tantamount to copyright infringement, potentially exposing AI companies to legal consequences. MEPs said that copyrighted material must be fairly remunerated to protect the EU's creative sector, which generates almost seven percent of the EU's GDP.
Under current EU rules, companies can use copyrighted materials for text and data mining, including AI training, unless a creator has reserved their rights. Publishers at Press Gazette's Media Strategy Network USA event agreed that where possible, unauthorized large language models should be blocked from training on their websites to pave the way for future deals. Some publishers drew a distinction between blocking AI access for training and allowing some retrieval-augmented generation to ensure publisher content still appears in real-time news-related results.
It is the height of insolence to just take others' life's work and grind it into a common brain and spit out what you want.
All publishers at the event appear to be using AI in-house in some form, but most agreed that writing articles should remain in the hands of human journalists. Concerns about falling referral traffic to publishers from Google were widespread at the event.
Artists have voiced strong criticism of AI services using their works without permission. Frode Øverli, creator of the Norwegian comic 'Pondus', criticizes AI services like Gemini for using artists' works without permission and calls for clearer regulations to protect copyright. The EU Parliament is aligned with Øverli on the need for clear regulations to protect artists' copyright from misuse by AI services.
The music industry has expressed similar concerns about reckless AI use of copyrighted material. Inger Elise May, director at Tono, the Norwegian music rights organization, says AI companies are acting recklessly by using music and culture in their models without permission or payment. Meanwhile, Erlend Hanstveit, State Secretary in the Norwegian Ministry of Culture, says the ministry has proposed several changes to copyright law in the digital market to the Storting, the Norwegian parliament, which could affect AI systems' ability to train on copyrighted material.
We understand that the cultural field is impatient and wants things to go faster.
In a separate development, the EU is banning employees from using AI-generated images and videos in official communication to strengthen trust in the union's institutions.
Several key unknowns persist in the evolving landscape of AI regulation and copyright protection. It remains uncertain whether Google's Gemini AI service violates copyright by generating content in the style of artists like Frode Øverli. The Norwegian government's specific plans to protect Norwegian culture from unauthorized use by AI companies have not been detailed publicly. The exact details and timeline for the proposed changes to Norwegian copyright law and their impact on AI training also await clarification.
Practical implementation challenges for publishers represent another area of uncertainty. How publishers will practically implement blocking AI training bots while allowing retrieval-augmented generation for real-time news results involves technical complexities that are still being worked out.