ASML, Mistral Ask EU to Delay Start of AI Act Rules

Mint

a day ago

(Bloomberg) -- European companies including ASML Holding NV and Mistral AI have called on the European Union to suspend the implementation of its landmark artificial intelligence regulation because they say it puts the continent's AI ambitions at risk.
In a letter signed by representatives from more than 45 organizations, business leaders asked the president of the European Commission, Ursula von der Leyen, to postpone the application of rules hemming in the most powerful AI models, calling for a more 'innovation-friendly regulatory approach.'
'This postponement, coupled with a commitment to prioritize regulatory quality over speed, would send innovators and investors around the world a strong signal that Europe is serious about its simplification and competitiveness agenda,' the letter said.
Other businesses to sign the letter include Airbus SE, Mercedes-Benz Group AG, BNP Paribas, Deutsche Lufthansa, Publicis Groupe SA, Prosus NV and Siemens Energy AG. They called for the delay to two sets of rules applying to general-purpose AI models and to high-risk AI systems.
Businesses are complaining that the commission has failed to deliver key guidelines and standards, including a code of practice which provides instructions for advanced AI companies to comply with the regulation. The drafting of the code of practice — already late after the commission originally slated it for May — has been dogged by criticism. Technology companies have accused it of going beyond the bounds of the EU's AI Act and creating its own set of onerous rules. The US government has also weighed in, and in April sent letters to the commission and a number of European countries that urged the bloc to scrap the code in its current form.
Now, these delays threaten to push back implementation of the act altogether. According to the EU's staggered deadline for the act, the most powerful AI models, such as OpenAI's ChatGPT, should be in compliance by August. But as of July the commission's working groups, made up of AI developers, academics and digital rights activists, were still hashing out detailed guidance.
The AI Act, passed last year, is a set of controls meant to prevent the worst abuses of the technology. AI developers will have to provide information about how their models are trained and have policies respecting copyright law. The most sophisticated models will have to take additional steps to mitigate risk, shore up security and report information about their system architecture. The act also puts guardrails on some types of use, such as restrictions on tools that would identify people in public spaces in real time.
While the code of practice is voluntary, it provides a framework that will help companies stay in compliance with the act. Breaking those rules can carry a fine of as much as 7% of a company's annual sales or 3% for the companies developing advanced AI models.
Speaking virtually at a Brussels event in February, Meta Platforms Inc.'s head of global affairs Joel Kaplan called the code of practice 'unworkable and infeasible,' adding that the company would not sign it in its present form. Alphabet Inc. has also criticized the rulebook and an executive said in an interview with Politico in February that guidelines calling for third-party model testing and requirements around copyright protection go too far.
(Corrects to remove SAP from the list of signatories, changes the number of people who signed)
More stories like this are available on bloomberg.com