A new stage of the European Union’s AI Act has come into force, introducing transparency requirements for artificial intelligence systems while delaying tougher rules governing high-risk AI applications until late 2027.

The legislation is widely viewed as the world’s first comprehensive AI regulatory framework and is expected to shape global AI governance much as the GDPR transformed privacy regulation.

Under Article 50, companies must clearly disclose when users are interacting with AI systems such as chatbots, while AI-generated content, including images, video, audio and text, must carry machine-readable identifiers where required.

Meanwhile, developers of biometric categorization and emotion-recognition systems must also inform affected individuals. Non-compliance can trigger fines of up to €15 million or 3% of worldwide annual turnover.

The EU postponed the Act’s high-risk AI obligations, which apply to systems used in sensitive areas including hiring, education, essential services, and migration, citing the need for additional technical standards and implementation guidance. Digital rights advocates criticized the decision, arguing it delays safeguards for vulnerable groups and rewards industry lobbying.

While the new transparency rules may become a de facto global standard, critics note that many AI-powered migration and border surveillance systems deployed outside of the EU remain beyond the Act’s jurisdiction.