Under the new rules, users must be clearly informed when they are interacting with an artificial intelligence system rather than a human, unless this is already obvious from the context, such as when using AI chatbots like ChatGPT.
The legislation also requires providers to label AI-generated or AI-manipulated images, audio recordings and videos, including so-called deepfakes. In addition to visible labels, AI-generated content must include technical markers, such as metadata embedded in files, so search engines and other digital services can automatically identify the material.
The new provisions also strengthen the powers of the European Commission's AI Office, which is now responsible for enforcing the rules for general-purpose AI models, including systems such as ChatGPT, Gemini and Claude. The office can launch investigations, request information and documentation, conduct inspections, and impose penalties if it determines that an AI system poses a risk or fails to comply with the regulation.
Authorities can also begin enforcing the AI Act's prohibitions on certain high-risk practices, including AI systems designed to manipulate or deceive users.
However, not all parts of the landmark legislation are taking effect at the same time. Amendments introduced under the EU's digital omnibus package, which entered into force on July 27, postponed several implementation deadlines.
Among the delayed measures is the requirement for member states to establish regulatory sandboxes for testing AI systems, which has been pushed back by one year to August 2, 2027. Some obligations for high-risk AI systems, including those used in recruitment and banking, have also been postponed until December 2, 2027.
Companies that violate the AI Act face fines of up to €15 million or 3% of their annual global turnover, whichever is higher, making the regulation one of the world's strictest legal frameworks governing artificial intelligence.
(aj)
SOURCE: PAP; IAR