Call for delay to EU rules on GPAI models and ‘high-risk’ AI

AI Act rules applicable to ‘general purpose’ AI (GPAI) models are due to take effect on 2 August 2025. A highly anticipated code of practice to support compliance with those rules, which was supposed to be published by the AI Office in early May this year, has yet to be finalised and published. Further AI Act rules applicable to ‘high-risk’ AI systems will take effect in August 2026.

Both sets of rules should be postponed, to allow for “innovation-friendly” changes to, and simplification of, the rules to be considered, a group of 50 European business leaders have said. They warned that the “Europe’s AI ambitions” are “at risk” as a result of “unclear, overlapping and increasingly complex EU regulations”. The group includes senior leaders from AI trade associations as well as companies such as Airbus, BNP Paribas, Mercedes Benz, and Phillips.

The group’s call, made in an open letter to senior EU officials, coincided with a report by Politico which suggested that the European Commission is considering delaying implementation of the GPAI code to the end of 2025.

According to Politico, the Commission will also decide by the end of August whether to delay implementation of the rules on ‘high-risk’ AI. That decision, according to Politico’s report citing an interview with EU commissioner Henna Virkkunen, will be influenced by whether new standards, being designed to support compliance with the ‘high-risk’ AI regime, have been finalised.

“We welcome recent discussions considering the need to postpone the enforcement of the AI Act as relevant guidelines and standards continue to be developed, and as various industries work together to find solutions that work for everyone,” the business leaders said.

“To address the uncertainty this situation is creating, we urge the Commission to propose a two-year ‘clock-stop’ on the AI Act before key obligations enter into force, in order to allow both for reasonable implementation by companies, and for further simplification of the new rules. This should apply both to obligations on e.g. high-risk AI systems, due to take effect as of August 2026, and to obligations for general-purpose AI models (GPAI), due to enter into force as of August 2025 while the corresponding and much anticipated code of practice has yet to be released,” they said.

“This postponement, coupled with a commitment to prioritise 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. In the context of the broader review of EU digital rules you have announced, it would also create the room needed to develop an innovation-friendly implementation strategy and identify pragmatic avenues for regulatory simplification, covering both GPAI models and high-risk AI systems as well as broader digital regulations. We have developed detailed proposals and are ready to work hand in hand with the Commission,” the group added.

“It is a remarkable letter”, said Dr. Nils Rauer, expert in AI law and regulation at Pinsent Masons. “It is important to note that the authors do not call into question the overall need for a sound regulatory framework for AI – their concern is legal certainty. AI, if used in a false manner, can become poisonous.”

“All new laws initially come with some level of uncertainty,” added Anna-Lena Kempf, also of Pinsent Masons. “There are no court decisions or administrative orders yet which could provide steer on how the legal provisions are to be understood and applied.”

EU proposals for “simplification” of the EU’s digital policy rulebook are expected to be published in the autumn. That Commission package is anticipated in response to increasing industry and political pressure to reduce regulatory burdens in digital markets. The US government, for example, has called on the EU to do more to support technological innovation by companies, with many US-headquartered technology companies operating in the EU market, while Mario Draghi, the former European Central Bank president, in a report into EU competitiveness, last year urged action to reduce regulation – particularly for tech companies.

Earlier this summer, EU law makers at the Council of Ministers were presented with wide-ranging proposals for digital regulatory reform. Those proposals, drafted by the Polish government, including potentially using ‘stop the clock’ legislative instruments to delay the effect of legislative provisions that have already been finalised – including the enforcement provisions in the AI Act – in the same way that has happened already in the context of EU sustainability-related due diligence and disclosure obligations.

Last month, the Commission open a consultation regarding implementation of the AI Act’s rules on ‘high-risk’ AI systems. According to the consultation, some changes to those rules are under Commission consideration – including in relation to the classification of high-risk AI systems and the obligations associated with providing, deploying, importing or distributing those systems. The consultation closes on 18 July 2025.

Rauer said: “It is no secret that the EU AI Act is a highly complex piece of legislation. The European legislator is trying to build a regulatory framework both for AI systems and AI models with distinguished obligations for providers, deployers, importers etc. In this context, it is a challenge for businesses to define their own role and to make sure their own products are compliant by design and by default.”

Out-Law revealed in April that the mainstream emergence of DeepSeek – the Chinese open source AI tool developed at a fraction of the cost of other LLMs on the market, and without access to the same computing power – has spurred discussions within the Commission around classification of GPAI models in the context of the scope of rules applicable to GPAI models under the AI Act.