The EU’s pioneering legislation to regulate AI is set to come into force next month, despite criticisms that it is incomplete, ambiguous, and stifling to the tech industry.
The first law of its kind anywhere in the world, the EU Artificial Intelligence Act aims to protect citizens from potentially harmful uses of AI by regulating companies within the EU, without losing ground to AI superpowers China and the US.
While EU lawmakers are “mostly concerned” about consumer safety and the dissemination of deepfakes, which could mislead voters in elections, the tech community has raised “gripes” with the legislation, said Euronews. Officials are now “frantically trying to plug the holes in the regulation before it comes into force”, said the Financial Times.
The act, which will be implemented in stages over the next two years, classifies different types of AI by risk.
Minimal risk uses – AI-powered video games and spam filters, for example – will not be subject to regulation. Limited risk activities, such as chatbots and other generative AI platforms, will be subject to “light regulation”, including transparency requirements to inform consumers that they are interacting with a machine.
The “high-risk” category includes AI systems used by law enforcement like biometric identification, as well as systems used to access public services or critical infrastructure. Developers of such systems will have until August 2027 to comply with rules around risk assessment and human oversight.
A further “unacceptable risk” category bans all AI systems that “threaten citizens’ rights”, said The Verge. This includes AI used to deceive or manipulate humans, or profile them as potential criminals based on behaviour or personality traits.
Progress on the legislation, which has been in the works for three years, was “upended” in 2022 when OpenAI released ChatGPT, said the FT. The emergence of “generative AI” models, which create text or images based on user prompts, “reshaped the tech landscape” and had parliamentarians “rushing to rewrite the rules” to regulate the large language models that underpin such apps.
“Time pressure led to an outcome where many things remain open,” a parliamentary aide involved in drafting the “rather vague” law told the FT. Regulators “couldn’t agree on them and it was easier to compromise” on what the unidentified aide called “a shot in the dark”.
Some critics say the text lacks clarity – especially on whether systems like ChatGPT are acting illegally when they use sources protected by copyright law. There is also confusion over who is responsible for content generated by AI, or what “fair remuneration” might look like for those who create the content it draws from. The act also does not specify who might enforce the rules in individual member states, or how, which could lead to patchy implementation across the continent.
The cost of compliance is also a problem, particularly for small companies. It could make it “very hard for deep tech entrepreneurs to find success in Europe”, Andreas Cleve, chief executive of Danish healthcare start-up Corti, told the FT. Many believe that cost will hinder those European companies competing with the US and China. “We will be hiring lawyers while the rest of the world is hiring coders,” said Cecilia Bonefeld-Dahl, director general for DigitalEurope, which represents the bloc’s technology sector.
Tech companies have until February next year to comply with rules on “unacceptable risk” content or face a fine of 7% of their total global annual revenue, or €35 million (£29.5 million), whichever is higher. Developers of the “high risk” category AI will have until August 2027 to comply.
By some estimates, the EU needs between 60 and 70 pieces of secondary legislation regarding the details of how the act will be implemented and enforced, and those must be in place by May next year. “The devil will be in the details,” a diplomat who took a leading role in drafting the act told the FT. “But people are tired and the timeline is tight.”