EU reaches deal to pass landmark AI law

One of the world’s first big efforts to regulate AI is nearing legislative reality.

After a marathon negotiation session, European Union lawmakers have agreed on a provisional deal to advance the sweeping AI Act. The landmark law is set to add new guardrails around everything from facial recognition to large language models (LLMs), and could serve as a blueprint for other countries looking to curb AI’s riskier tendencies.

Negotiators for the EU Parliament, EU executives, and leaders of its 27 member states came to the agreement after more than 30 hours of discussion, according to EU Commissioner Thierry Breton. Legislators will next hash out the bill’s final wording before it is passed by the Parliament and the Council of country leaders, a vote that one Parliament member told the Associated Press is more of a formality now that the deal has been reached.

“Europe has positioned itself as a pioneer, understanding the importance of its role as a global standard setter,” Breton said in a press conference. “This is, I believe, a historic day.”

Rules all the rage: The rise of generative AI has spurred governments all over the world to look at how best to control against the tech’s worst downsides, which can range from job replacement to biased hiring systems to humanity’s doom, depending on whom you ask.

In October, the Biden administration rolled out an executive order taking aim at AI; a month later, the United Nations convened a body to study global AI rules and world leaders congregated at a summit in England to sign an agreement that regulating the tech would take international cooperation.

Among the various regulatory efforts, the AI Act stands out as one of the most far-reaching, substantive, and meaningful in terms of carrying the force of law.

Enforcement learning: The scope of the AI Act spans areas like facial recognition and other biometric data collection and data transparency around LLMs. It makes certain exceptions for police uses, which were reportedly a sticking point in the hours of debate.

Under the law, AI companies will be obligated to disclose to consumers whether content is AI-generated. Certain high-risk AI systems will be subject to more rigorous development and testing rules, including disclosure of data used in training and statements addressing how the systems will impact fundamental rights. AI used in the insurance and banking sectors would also be subject to these impact statements.

Despite the historic nature of the agreement, some rights advocates said the bill did not go far enough toward keeping the tech in check. Sarah Chander, senior policy adviser at advocacy group European Digital Rights, said that while the law includes some important milestones, it still lacks some key protections.

“Whilst the deal contains some limited gains for human rights, it will be mostly a shell of the AI law Europe really needs,” Chander said.