Al Jazeera analyzes EU AI Act's delayed high-risk rules taking effect this week

An Al Jazeera explainer examines which parts of the EU Artificial Intelligence Act began to apply on 2 August and which were postponed. It says the Act is meant to complement existing EU digital rules, and asks whether it will shape AI governance as the GDPR shaped privacy.

What took effect is Article 50, a transparency layer. Chatbots and similar systems dealing directly with people must reveal that they are AI, unless that is plain. Providers of systems that generate or alter images, audio, video or text must make synthetic content identifiable through machine-readable marking where required. Systems that recognise emotions or categorise people by biometric data must tell individuals. Breaches can bring fines of up to 15 million euros or 3 percent of worldwide yearly turnover if that is larger. The article calls the impact on companies mostly operational, since no prior approval is needed, though firms must find where AI is embedded, including in vendor tools.

The high-risk obligations, for fields like biometrics, jobs, schooling, essential services and migration and border control, were due on the same date but were postponed in May to 2 December 2027 under the Digital Omnibus package. Henna Virkkunen said the aim was to ease innovation without lowering safety standards, and the Commission tied the change to its competitiveness agenda. Digital rights groups say the delay weakens protections. Stefi Richani of the Equinox Initiative for Racial Justice says it will increase surveillance and discrimination affecting migrants, and favours banning such systems.

The article adds that transparency standards may spread worldwide, but EU-funded border technology in third countries falls outside the Act. Earlier stages covered prohibitions and AI literacy in February 2025 and general-purpose model duties in August 2025.