Brussels: The European Union (EU) announced the entry into application of AI Act obligations for providers of general-purpose AI (GPAI) models across the EU, aiming to bring "more transparency, safety and accountability to AI systems on the market." EU Commissioner for Tech Sovereignty, Security and Democracy, Henna Virkkunen, emphasized the importance of these rules by highlighting the everyday reliance on General Purpose AI models, stating that the new obligations will make these models more transparent, trustworthy, and safe to use. The statement detailed that the rules are designed to ensure clearer information about AI model training, better enforcement of copyright protections, and accountable AI development.
According to Kuwait News Agency, the Commission has published guidelines to assist providers in complying with the new rules, including a standardized template for summarizing the data used to train the models. GPAI models are characterized by being trained with over 10^23 FLOP and capable of generating language. Moreover, the GPAI Code of Practice, developed by independent experts, is recognized as a voluntary tool for demonstrating compliance with the AI Act, offering providers who adhere to it reduced burden and increased legal certainty.
Providers of GPAI models are now required to meet transparency and copyright obligations when placing their models on the EU market. Existing models must ensure full compliance by August 2, 2027. Advanced or impactful models exceeding 10^25 FLOP, posing systemic risks, will face additional obligations, including notifying the Commission and ensuring the models' safety and security.
A total of 26 companies, including major tech firms such as Amazon, Google, Microsoft, and IBM, have signed the Code of Practice on General-Purpose AI Models. AI providers like OpenAI, France's Mistral AI, and Germany's Aleph Alpha are also signatories. However, Meta has opted out, arguing that the Code restricts innovation and that Europe's AI approach is misguided.
The EU's 27 member states are mandated to appoint national oversight authorities to monitor AI Act implementation and ensure compliance. Companies violating the rules may face fines of up to 17 million dollars or 3% of their annual global turnover.