The European AI Act: here are the obligations already in force for businesses and professionals
The Artificial Intelligence Regulation postpones certain deadlines, but already requires training, transparency and oversight
With the European Council’s final approval on 29 June 2026 of the simplification package for AI Act, Europe has opted to ease the timetable for the regulation’s implementation, postponing many of the most demanding obligations relating to ‘high-risk’ artificial intelligence systems until 2 December 2027.
However, this postponement should not be interpreted as a suspension of the AI Act. The European decision does not, in fact, remove the obligations already in force, nor those that will become central from August onwards for businesses, SMEs, professional practices and organisations that use artificial intelligence tools in their processes. The official European timeline itself confirms that the regulation will be applied progressively: from 2 February 2025, the requirement for AI literacy – amongst other provisions – will already be in force, as of 2 August 2026, the transparency rules and the strengthening of the supervisory framework will come into force.
The first requirement that should not be underestimated is already in force: AI literacy. Article 4 of the AI Act requires professional users of AI systems to take measures to ensure that their staff and anyone using such systems on their behalf have a sufficient level of competence. For an SME or a professional practice, this does not necessarily mean organising courses or obtaining formal certifications. It does, however, mean being able to demonstrate that those using AI within the organisation have been equipped to do so in an informed manner. Consider, for example, a small business that uses a chatbot to prepare commercial quotations, or a professional practice that uses AI tools to summarise documents, prepare drafts or analyse client materials. In all these cases, it is advisable to have internal guidelines, training sessions, rules on what data should not be included in prompts, criteria for verifying outputs, and clear instructions on human oversight.
What’s new from 2 August
From 2 August 2026, following a transitional period lasting approximately two years, Article 50 of the AI Act will come into full force. This stipulates that, when a person interacts directly with an AI system, they must be informed of this fact, unless it is evident from the context. This requirement applies to virtual assistants, customer care systems and automated interfaces on websites: if a service provider or e-commerce business installs a chatbot on its website to respond to users, it must make it clear that the respondent is not a person.
Article 50 itself introduces obligations relating to content generated or altered using AI. Systems that generate audio, images, video or text must ensure that the output is recognisable as artificial, including through machine-readable markings. For those who use these tools in communications, this applies to promotional images, renderings, videos, social media content, newsletters and informational content produced or reworked using AI. For example, an estate agency that publishes AI-generated images to depict a property, an architectural practice that creates renderings, or a communications agency that produces visuals for clients will need to use software which, when generating content, allows it to be downloaded with the appropriate metadata and combines the technical labelling with a visible statement on the page or in the caption.
