Three proposed changes to the AI Act, the EU’s main law on artificial intelligence, touch hiring software, image generators and the fines smaller firms can face. The European Parliament’s package had cleared committee stage but was not final, with a vote of the full chamber set for 26 March 2026. The law reaches beyond Europe, too. Under Article 2, it can apply to companies outside the EU that supply or use an AI system, when that system’s output is used in the Union.
The first change is timing. Lawmakers backed delaying some rules for high-risk systems, meaning AI used in sensitive areas where errors can seriously affect people’s rights or access to key services. The delay covers systems listed directly in the law, including uses in education, employment, essential services, law enforcement, justice and border management.
Hiring software shows what that means in practice. The law already treats recruitment tools, candidate filtering, promotion decisions, dismissal support, task allocation and worker monitoring as high-risk, because they can affect careers, income and workers’ rights. That classification stays, and only the timing would change.
A ban comes second. The package would outlaw AI tools that create or manipulate sexually explicit or intimate images of an identifiable real person without that person’s consent. These are known as nudifier systems. The proposal targets the capability itself, so a general-purpose image generator is covered too, unless effective safeguards stop it from producing such images.
The third change concerns company size. Parliament backed extending support measures to small mid-cap enterprises, companies larger than small and medium-sized firms but smaller than large corporate groups. Under the European Commission’s definition, these are firms with fewer than 1,500 employees. They must also have annual turnover of up to €300 million or a balance sheet total of up to €150 million.
Penalties would follow the same logic. The Parliament report would amend Article 99, the law’s penalties section. Small mid-caps would then face fines up to the lower of the two limits the law sets, a percentage-based amount or a fixed amount. That lighter treatment would not apply to providers of very large general-purpose AI models, whose failure or misuse could affect many markets at once.
