EU questions dozens of companies using new AI powers

The European Union (EU) said on Tuesday it had begun to use new powers to regulate artificial intelligence (AI) by sending information requests to dozens of companies after incidents this summer highlighted the risks the technology poses.

The European Commission gained new regulatory powers last month under the AI Act, the world's most far-reaching rules to curb potentially dangerous uses of the technology. It entered into force in 2024, with obligations kicking in over several years.

The Commission said on Tuesday it had sent requests for information to more than 30 companies operating in the AI sector.

The requests are a preliminary step before the possible launch of a formal investigation into whether companies are respecting European laws.

"Our goal is to ensure that AI in Europe is developed, released and used safely and transparently," Henna Virkkunen, the Commission's vice president for tech sovereignty and security said in a LinkedIn post over the weekend.

"And we are ready to take all necessary steps to ensure that companies comply with their obligations under the AI Act," she added.

The Commission did not identify the companies to which it sent letters, although Virkkunen said in her post they are located across the world.

Commission spokesman Thomas Regnier said the questionnaires concerned mostly safety and respect of copyright.

A series of incidents in recent months highlighted the risks the most advanced AI models now pose.

ChatGPT maker OpenAI acknowledged in July that its advanced AI models had gone rogue during security testing, hacking on their own into a popular platform for programmers.

Rival Anthropic also said in July that its AI models had "gained unauthorized access" to three outside organizations during testing that was supposed to keep them away from "real-world" systems.

When questioned about the incidents, Regnier said the Commission takes such events very seriously and maintains close contacts with the companies, but did not provide any details about their discussions.