Berliner Tageblatt - EU unveils recommendations to rein in powerful AI models

Brussels has come under fierce pressure to delay enforcing its landmark AI law as obligations for complex models known as general-purpose AI -- systems that have a vast range of functions -- kick in from August 2.

The law entered into force last year but its different obligations will apply gradually.

But as the EU pivots to bolstering its competitivity and catching up with the United States and China, European tech firms and some US Big Tech want Brussels to slow down.

The European Commission, the bloc's digital watchdog, has pushed back against a delay.

The EU's executive arm has now published a code of practice for such systems prepared by independent experts with input from others including model providers themselves.

In the code, the experts recommend practical measures such as excluding known piracy websites from the data models use.

The code applies to general-purpose AI models, such as Google's Gemini, Meta's Llama and X's Grok -- the tech billionaire Elon Musk's chatbot that has come under fire this week for antisemitic comments.

Under the law, developers of such models must give details about what content they used -- like text or images -- to train their systems and comply with EU copyright law.

The code was due to be published in May. EU officials reject claims that it had been watered down in the past few months due to industry pressure.

Corporate Europe Observatory and Lobby Control in April had accused Big Tech of "heavily" influencing the process "to successfully weaken the code".

The code will need endorsement by EU states before companies can voluntarily sign up to it.

Businesses that sign the code "will benefit from a reduced administrative burden and increased legal certainty compared to providers that prove compliance in other ways", the commission said in a statement.

Nearly 50 of Europe's biggest companies including France's Airbus, Dutch tech giant ASML and Germany's Lufthansa and Mercedes-Benz last week urged a two-year pause.

The companies' CEOs in a letter accused the EU's complex rules of putting at risk the 27-country bloc's AI ambitions and the development of European champions.

The EU will be able to enforce the rules for general-purpose AI models a year from August 2 for new models, while existing models will have until August 2027 to comply.

L.Janezki--BTB