European Union Unveils Rules for Powerful A.I. Systems

European Union officials unveiled new rules on Thursday to regulate artificial intelligence. Makers of the most powerful A.I. systems will have to improve transparency, limit copyright violations and protect public safety.

The rules, which are not enforceable until next year, come during an intense debate in Brussels about how aggressively to regulate a new technology seen by many leaders as crucial to future economic success in the face of competition with the United States and China. Some critics accused regulators of watering down the rules to win industry support.

The guidelines apply only to a small number of tech companies like OpenAI, Microsoft and Google that make so-called general-purpose A.I. These systems underpin services like ChatGPT, and can analyze enormous amounts of data, learn on their own and perform some human tasks.

The so-called code of practice represents some of the first concrete details about how E.U. regulators plan to enforce a law, called the A.I. Act, that was passed last year. Tech companies played a major role in drafting the rules, which will be voluntary to adopt when they take effect on Aug. 2. E.U. regulators will not be able to impose penalties for noncompliance until August 2026, according to the European Commission, the executive branch of the 27-nation bloc.

The European Commission said the code of practice is meant to help companies comply with the A.I. Act. Companies that agreed to the voluntary code would benefit from a “reduced administrative burden and increased legal certainty.” Officials said those that do not sign would still have to prove compliance with the A.I. Act through other means, which could potentially be more costly and time-consuming.

It was not immediately clear which companies would join the code of practice. Google and OpenAI said they were reviewing the final text. Microsoft declined to comment. Meta, which had signaled it will not agree, did not have an immediate comment. Amazon and Mistral, a leading A.I. company in France, did not respond to a request for comment.