EU AI Act deadline shifts focus to transparency and enforcement

The Digital Omnibus has postponed the AI Act’s high-risk obligations under Annex III from 2 August 2026 to 2 December 2027, while rules for AI embedded in regulated products move to 2 August 2028. The delay reflects concerns that harmonized standards and notified bodies were not ready, but it does not pause the wider compliance timetable.

Several obligations still take effect on 2 August 2026 across the EU. Article 50 transparency rules will require clear disclosures for chatbots and virtual assistants, labels for deepfakes and synthetic public content, notices for emotion recognition and biometric categorization, and machine-readable marking for generative systems. Systems already on the market before August 2026 receive a grace period until 2 December 2026 for machine-readable marking.

Enforcement also becomes more significant as Article 99 penalties activate, including fines of up to €35 million or 7% of worldwide turnover for prohibited practices, up to €15 million or 3% of worldwide turnover for other breaches, and up to €7.5 million or 1% of worldwide turnover for incorrect information supplied to authorities. Businesses are advised to inventory AI systems, classify risk, deploy transparency notices, document staff training, and prepare early for high-risk requirements.