Why the EU AI Code is Splitting Top AI and Tech Leaders

The EU’s AI Code of Practice has exposed sharp divisions among technology companies, with some players embracing the framework while others reject it as regulatory overreach.

The code serves as a practical implementation guide for the EU AI Act, the world’s first AI legislation that aims to regulate AI systems based on their potential risks. 

While the act establishes a risk-based framework that imposes stricter requirements on AI systems deemed to pose greater threats to safety, fundamental rights and society – the voluntary Code of Practice specifically addresses general-purpose AI models.

It addresses general-purpose AI models because they are powerful systems like those developed by OpenAI, Google and Meta that can perform multiple tasks rather than being designed for specific applications. 

These models, which include large language models (LLMs) and multimodal systems, are some of the most advanced AI technologies currently available.

Now, major technology and AI companies have started to sign or reject the code.

Who is agreeing and who is rejecting?

The voluntary code was developed by 13 independent experts with input from more than 1,000 stakeholders, including model providers, small and medium-sized enterprises, academics, AI safety experts, rights-holders and civil society organisations.

Yet the split between technology and AI companies comes as the code’s August 2025 deadline approaches for mandatory compliance with the EU AI Act.

Microsoft
​​​​​​​Microsoft is likely to sign.

“I think it’s likely we will sign. We need to read the documents,” says Brad Smith, Microsoft’s Vice Chair and President.

“Our goal is to find a way to be supportive, and at the same time, one of the things we welcome is the direct engagement by the AI Office with industry.”

OpenAI and Mistral
​​​​​​​OpenAI, has already committed to signing the code, positioning itself as an early adopter of the voluntary framework.

The company also joins French AI developer Mistral as signatories to the agreement.

“Signing the code reflects our commitment to providing capable, accessible and secure AI models for Europeans to fully participate in the economic and societal benefits of the Intelligence Age,” OpenAI says in its announcement.

Antropic
​​​​​​​Anthropic has additionally announced its intention to sign the code.

“We believe the code advances the principles of transparency, safety and accountability – values that have long been championed by Anthropic for frontier AI development,” the company says.

Meta
​​​​​​​Meta’s position however, stands in stark contrast. 

Joel Kaplan, the company’s Chief Global Affairs Officer, argues that “Europe is heading down the wrong path on AI” and warns the code would “throttle the development and deployment of frontier AI models in Europe and stunt European companies looking to build businesses on top of them”.

Joel, who replaced Nick Clegg earlier this year and previously served as Vice President of US policy at Facebook, writes that “the code introduces several legal uncertainties for model developers, as well as measures which go far beyond the scope of the AI Act”.

The EU AI Code’s three key requirements

The code establishes requirements across three areas for companies developing general-purpose AI models:

• Transparency
• Copyright compliance
• GPAI with Systemic Risk (GPAISR)

Transparency obligations require providers to maintain technical documentation about their model and datasets. 

While copyright compliance mandates clear internal policies outlining how training data is obtained and used under EU copyright rules.

For the most advanced models, classified as “GPAI with Systemic Risk” (GPAISR), additional safety and security obligations apply. 

This category covers the most powerful AI systems, including OpenAI’s o3 model, Anthropic’s Claude 4 Opus and Google’s Gemini 2.5 Pro.

Furthermore, signatories must publish summaries of content used to train their models and implement robust risk assessments. 

The framework requires companies to establish governance frameworks for managing potential threats from AI systems.

The penalties for non-compliance are substantial – as companies face fines of up to €35m US$38.5m or 7% of global annual turnover, whichever is greater. 

For providers of general-purpose AI models specifically, the European Commission may impose fines of up to €15m US$16.5m or 3% of worldwide annual turnover.

Industry pushback grows ahead of deadline

More than 40 of Europe’s largest businesses, including aerospace manufacturer Airbus and semiconductor equipment maker ASML Holding, have signed a letter asking the European Commission to delay implementation of the AI Act by two years.

As a result, the Commission has indicated that companies adhering to an approved Code of Practice will face a simplified compliance path. 

Regulators will also focus enforcement on checking that code commitments are met rather than conducting comprehensive audits of every AI system.

This creates incentives for early adoption among companies seeking regulatory predictability as mandatory enforcement begins in August 2025. 

Meanwhile, under the AI Act, obligations for general-purpose AI models become enforceable twelve months after the legislation enters force on 2 August 2025.

The regulatory framework aligns with broader global AI governance developments, including the G7 Hiroshima AI Process and various national AI strategies, potentially establishing European approaches as international benchmarks for AI regulation.

Despite mounting opposition, former Internal Market Commissioner Thierry Breton has insisted the framework will proceed as scheduled.