AI and the AI Act 2026: what changes for businesses

Quick answer. The AI Act (EU Regulation 2024/1689) has regulated AI across the European Union since 1 August 2024, with a phased application running to 2028. For a French business, the 2026 stake is to know whether your uses fall under prohibited risk, high risk or simple transparency, then to organise AI literacy, documentation and human oversight before the deadlines.

Artificial intelligence has settled into companies' everyday tools, from chatbots to recruitment software. In parallel, the European Union has adopted the world's first comprehensive legal framework on AI. Many business owners believe this text only targets technology giants. That is a mistake: it also concerns SMEs that merely use AI. Here is what changes concretely, and how we support owners in their digital transformation.

Regulation (EU) 2024/1689, known as the AI Act, was adopted on 13 June 2024 and published in the Official Journal of the Union on 12 July 2024. As a regulation, it applies directly across the 27 Member States, without national transposition. It entered into force 20 days after publication, on 1 August 2024, but its obligations roll out in stages through 2028.

The text targets two main actors: the provider, who develops or places an AI system on the market, and the deployer, the business that uses an AI system in the course of its activity. The vast majority of French SMEs are deployers: they do not write the algorithm, they use it. This does not exempt them from every obligation, as we will see. The scope is also extraterritorial: a vendor established outside the Union whose system is used in France falls within the regulation.

The AI Act classifies uses, not technologies. The same model can be harmless in one case and high risk in another. Four levels structure the regulation.

For an SME, the level that surprises most is high risk, because it covers uses that are very common in human resources. Software that automatically screens applications or evaluates employees falls under Annex III of the regulation: it is a high-risk use, with heavy obligations attached.

Obligations do not enter into force all at once. Here are the milestones set by the text, as amended by Regulation (EU) 2026/1744.

Between publication and full application of the high-risk strand, more than three years now pass. This delay is not wasted time: it serves to map your uses, train your teams and document your systems. Many businesses discover late that an HR tool or a scoring module bought off the shelf falls under high risk.

The distinction is decisive, because the obligations do not weigh in the same way.

The provider of a high-risk system bears the heaviest load: a risk management system, complete technical documentation, logging, CE marking, declaration of conformity, and registration in the European database.

The professional deployer has more targeted but very real obligations: use the system in line with its instructions, ensure effective human oversight, monitor operation, keep the logs for the required period (at least 6 months in several cases), and inform the persons concerned when a decision affecting them relies on a high-risk system.

Three obligations deserve a business owner's attention this year.

AI literacy (Article 4) has been in force since 2 February 2025. Providers and deployers must ensure a sufficient level of AI understanding among the people who use it. In practice, this means awareness and training tailored to the company's real uses, as shown in our article on AI agents to automate the back office.

Transparency (Article 50) requires, from 2 August 2026, informing the user that they are interacting with an AI and labelling content generated or manipulated by AI (text, image, sound, video). A chatbot on your site or visuals produced by AI are directly concerned.

Human oversight will apply to any high-risk use, from 2 December 2027 for Annex III and from 2 August 2028 for Annex I: a person must be able to understand, monitor and, if needed, override the automated decision. On a CV screening tool, this means a human keeps control over the final selection.

The regulation sets some of the highest fines in EU law (Article 99).

The cap applied is, in principle, the higher of the fixed sum and the percentage. For SMEs and start-ups, however, the regulation provides that the fine corresponds to the lower of the two, so as not to penalise small structures disproportionately.

The timeline has been adjusted. On 7 May 2026, the Council, the Parliament and the Commission reached a provisional political agreement, known as the Digital Omnibus, which became Regulation (EU) 2026/1744 of 8 July 2026 and defers the application of Annex III high-risk obligations from 2 August 2026 to 2 December 2027, and those of Annex I to 2 August 2028. The text also introduces new bans targeting AI-generated child sexual abuse material and non-consensual intimate imagery, applicable from 2 December 2026.

This regulation was published in the Official Journal of the Union on 24 July 2026 and has been in force since 27 July 2026. We recommend continuing to prepare for compliance on the basis of this timeline. Anticipating almost always costs less than catching up.

In our files, the AI Act raises less a technological question than a governance one. The first useful reflex is not legal but documentary: list the AI tools actually used in the business, which often holds surprises (a scoring module in the CRM, application screening in the HR software, a drafting assistant). Once the map is laid out, classification by risk level becomes simple, and the compliance burden narrows to two or three uses at most.

This is especially true for tech start-ups that embed AI at the core of their product, and that may be a provider on one component and a deployer on another. For the accounting and finance function itself, we have detailed the uses, the return on investment and the risks in our guide to AI in accounting. The AI Act must not slow AI adoption: handled well, it remains a productivity lever, as shown by our generative AI use cases for a business owner.

A 40-employee services SME asks us about its recruitment software, which automatically ranks applications by relevance score. This use falls under Annex III: it is high risk. The company is a deployer, not a provider, but it must still ensure human oversight of the selection, inform candidates that the tool is involved, and keep the logs. We helped the owner obtain the conformity documentation from the vendor, build human oversight into the HR process, and add an information notice in the candidate journey. The work took a few weeks, without blocking the activity, because it was handled before the deadline rather than in a rush.

For the related issues of data security, also see our cybersecurity checklist for SMEs and our note on the NIS2 directive.