The AI Act (Regulation (EU) 2024/1689) assigns duties to roles: providers, who develop AI systems and place them on the market under their own name, and deployers, who use AI systems under their own authority (Art. 3(3) and (4)). Almost every organisation is a deployer today – through chat assistants, AI features in office software, translation services or recruiting tools. The duties fall on the organisation as a whole. Organising how they are met is therefore a job for leadership.
Some duties already apply. Since 2 February 2025 certain AI practices are prohibited (Art. 5), and providers and deployers must take measures on the AI literacy of their staff (Art. 4). Since 2 August 2026 the transparency duties in Art. 50 apply, for example to chatbots and AI-generated content. The Digital Omnibus (Regulation (EU) 2026/1744, in force since 27 July 2026) rewrote Art. 4: it now requires measures to promote AI literacy rather than ensuring a particular level.
The most extensive duties concern high-risk AI – systems in the areas listed in Annex III, such as candidate selection, employee performance assessment, creditworthiness checks or access to essential services. After the Digital Omnibus these duties apply from 2 December 2027, and for AI in regulated products under Annex I from 2 August 2028. That sounds far off, but it is tight for building an inventory, roles and evidence if you only start in 2027.
For leadership it comes down to three questions. First: do we know which AI we use and for what? Second: is it clear who is responsible for what and who decides? Third: can we prove it – to supervisory authorities, customers, auditors and our own supervisory board? An organisation that can answer yes has done most of the work before the strictest duties kick in.
This guide deliberately avoids horror scenarios. The maximum fines are real, but they are ceilings. For most organisations the bigger risk is a different one: projects stopped because questions about classification, the works council or data protection were asked too late, and customer questionnaires that nobody can answer with confidence.