EU AI compliance guide may be delayed until late 2025, commission says

A long-awaited code of practice to guide businesses on EU AI rules may not take full effect before late 2025, leaving companies facing legal uncertainty as enforcement deadlines approach.

The code of practice, which will apply to large language models such as OpenAI’s GPT-4 and competing systems developed by Google and Mistral, was initially scheduled for release in early May.

A Commission spokesperson has now said the non-binding guidance will be published in the coming days, with companies invited to sign on voluntarily starting in August, and the code potentially taking effect by year-end, according to Reuters.

While the code is voluntary, the EU AI Act’s rules for general-purpose AI models will become legally binding from August 2, with enforcement staggered over the following two years.

Several major tech companies have called for a delay in enforcing the AI Act, citing a lack of practical guidance on implementation as a key concern.

This week, more than 40 CEOs from leading European companies, including ASML, Philips, Siemens, and Mistral, appealed to European Commission President Ursula von der Leyen to pause the rollout of the AI Act for two years.

In a letter, the executives warned that “unclear, overlapping and increasingly complex EU regulations” threaten to undermine Europe’s AI goals, putting both the growth of homegrown tech firms and broader industry adoption at risk amid rising global competition.

High-stakes compliance

The delay in publishing the guidelines carries weight, as the document is expected to offer much-needed clarity for companies seeking to navigate the complex requirements of the AI Act.

“Since the maximum fine for breaking the rules of the AI Act is 7% of global revenue, the stakes are high,” said Hyoun Park, CEO and chief analyst at Amalgam Insights. “CIOs responsible for either following or tracking activity associated with the EU AI act need to carefully check whether the models they are using actually provide this level of lineage and visibility.”

Uncertainty over what the regulations will ultimately permit is also raising concerns that vendors may be forced to slow down AI deployments.

“Vendors liable for following the AI Act may end up having to move more slowly in Europe, as they are unable to fully parse the AI Act or the activities that are allowed,” Park added.  

CIOs on alert

With the rules set to kick in soon, analysts say enterprise technology leaders must act now to ready their systems, vendors, and processes for a rapidly evolving regulatory landscape.

“AI has several stakeholders, and getting them all on a common page will not be easy,” said Faisal Kawoosa, founder and lead analyst at Techarc. “It will take time until the interests of all, especially big tech companies, are safeguarded. While implementations will continue in the enterprise environment, I think CIOs will have to be prepared for compliance checks and re-checks.”

The eventual adoption of the code may not upend how companies currently use AI, but it will introduce new layers of scrutiny, Kawoosa added.

Others see the delay in guidance as a strategic opportunity for CIOs to reinforce internal processes before enforcement begins.

“CIOs should view the delay as a window to strengthen internal governance and vendor assessments,” said Abhishek Sengupta, practice director at Everest Group. “It’s a signal to not pause deployments, but to prepare for future compliance by ensuring transparency and accountability mechanisms are in place today.”EU AI compliance guide may be delayed until late 2025, commission says – ComputerworldRead More