Imagine a company using AI to screen job applications, or a manufacturer building AI into a machine. If the system makes a wrong decision, the result may affect more than a search or recommendation. It could change someone’s access to work, education or essential services, or even put people’s safety at risk. This is why the EU AI Act uses a risk-based approach: the greater the potential harm, the stronger the safeguards.
On 29 June 2026, the Council of the European Union adopted the AI Omnibus. It entered into force on 27 July. The AI Omnibus is not a second AI Act. It amends the existing law and gives businesses, public authorities and regulators more time to prepare. The most visible change is the delayed application of some high-risk AI rules. For stand-alone systems used in sensitive areas such as recruitment, education, public services and biometrics, the rules will apply from 2 December 2027. For high-risk AI built into products such as machinery, toys and lifts, they will apply from 2 August 2028.
A later deadline does not mean that high-risk AI is free from oversight. AI providers and developers should use the extra time to review data quality, assess risks, establish human oversight and cybersecurity controls, and prepare technical documentation and activity records. Manufacturers that embed AI into products should also examine how product-safety rules and AI requirements fit together. The preparation period is longer. The responsibility remains.
The AI Omnibus also draws a clearer red line. AI systems must not generate non-consensual sexually explicit or intimate content, including AI nudification tools and other sexual deepfakes. They must not generate child sexual abuse material either. These new prohibitions will apply from 2 December 2026.
For individuals, the useful question is not only “Is this image real?” but also “Did the person shown consent to it?” For companies, developers, product manufacturers and public authorities, the task is to build compliance into design, procurement and daily operations rather than treat it as a repair job after harm occurs. The AI Omnibus gives some high-risk AI rules more time to arrive. It does not create room for violations of privacy, dignity or child safety.