New Deadlines for AI: Companies Given More Time, Bans Introduced

Members of the European Parliament today agreed in a Plenary session in Brussels on proposals to simplify the rules on artificial intelligence, the so-called digital omnibus, which includes a ban on ‘nudification’ systems using AI and a postponement of the application of certain rules for high-risk systems.

With 569 votes in favor, 45 against, and 23 abstentions, the EP adopted a position that amends the Artificial Intelligence Act (AI Act), and in its amendments, the representatives introduced specific dates for the application of the AI Act to ensure predictability and legal certainty.

For high-risk artificial intelligence systems, namely those involving biometrics and those used in critical infrastructure, education, employment, essential services, law enforcement, justice, and border management, the Members of the European Parliament propose the start of the application of the AI Act on December 2, 2027.

For artificial intelligence systems covered by EU sectoral legislation on safety and market surveillance, August 2, 2028, has been proposed as the first day of application.

Additionally, service providers are given a deadline until November 2, 2026. to comply with the rules on watermarking audio, visual, video, or textual content created by artificial intelligence to indicate its origin.

Furthermore, a new ban is introduced on artificial intelligence systems that can create sexually explicit images and manipulate them (deepfakes) that resemble a recognizable real person without that person’s consent. However, the ban does not apply to AI systems with effective security measures that prevent users from creating such images.

Although MEPs supported allowing service providers to process personal data to detect and correct bias in artificial intelligence systems, they introduced safeguards to ensure that this is done only when explicitly necessary.

To prevent overlap in the application of EU sectoral rules on product safety and the Artificial Intelligence Act, obligations under the Artificial Intelligence Act may be less stringent for products already regulated by sectoral laws (e.g., medical devices, radio equipment, toy safety, etc.).

Finally, to help EU companies expand as they outgrow their status as small and medium-sized enterprises (SMEs), where they enjoy certain support measures, representatives supported the proposed extension of these measures to small companies of medium market capitalization.

Following this decision by Parliament, negotiations with the Council on the final form of the AI Act will follow, which should be completed by the end of April this year, explained German MEP Axel Voss from the European People’s Party, who worked on the legal framework for artificial intelligence.

– The key thing about the deadlines in the AI Act is ‘stopping the clock’ so that companies do not have to panic to be ready by August 2. This gives them more time. This is a new law, and no one actually knows how to deal with it yet, so it is justified that companies get additional time to prepare, Voss explained.

Voss believes that Parliament is successfully dealing with this challenge, but emphasizes that the current proposals will not change the whole world as some formulations in the AI Act remain vague. According to him, it is essential to develop clear and practical guidelines that will serve as a roadmap for companies in their operations.

He also acknowledges that there are still many elements that can be improved, but emphasizes that this was unfeasible under the current time pressure. It is now expected that the European Commission’s AI Office will come up with concrete templates and guidelines that will facilitate the application of the law in practice.

– I am very optimistic that we will successfully conclude negotiations by the end of April, after which translations of the law will follow, a re-vote in the plenary session, and publication in the EU’s official journal, Voss concluded.