While individuals can sue AI providers under the new law, enforcement by the European AI Office will only begin later. — REUTERS/Yves Herman//File Photo
BRUSSELS: New transparency rules for general-purpose artificial intelligence (AI) models such as ChatGPT and Gemini came into effect across the European Union on Saturday, as part of the bloc's landmark AI Act adopted last year.
Under the new rules, developers must disclose how their models work and what data was used to train them. Particularly advanced models, seen as posing potential risks to the public, are also required to document safety measures.
General-purpose AI refers to systems that can be used for a wide range of purposes, such as generating text, analysing language, or writing code.
The legislation aims in part to bolster copyright protections. Developers must now report the sources of their training data and whether they used automated web scraping. They are also required to outline steps taken to safeguard intellectual property and establish contact points for rights holders.
However, several associations representing authors, artists and publishers have criticised the law for not going far enough. In a joint statement, the German-based Initiative Urheberrecht, a copyrights advocacy group, argued the rules remain ineffective without a requirement to name specific datasets or domains.
While individuals can sue AI providers under the new law, enforcement by the European AI Office will only begin later. The office will supervise new models from August 2026 and models released before August 2, 2025 starting in August 2027.
Violations could result in fines of up to €15mil (RM73.2mil) or 3% of global annual sales.
To ease compliance, the European Commission also issued voluntary guidelines and a code of conduct. Google, developer of the Gemini model, said it intends to sign the code, though it warned the regulation could stifle innovation. – dpa
