Meta has refused to sign the European Union's code of practice for its AI Act, challenging the EU's regulatory approach to artificial intelligence. The company's chief global affairs officer, Joel Kaplan, cited legal uncertainties and measures exceeding the AI Act's scope as reasons for their decision. The code, published in July, is a voluntary framework to help companies comply with the EU's AI legislation, including provisions on AI tool documentation and training data restrictions. Meta's stance reflects broader concerns about the EU's regulatory approach, with the company viewing it as potential overreach that could hinder AI development in Europe. This decision puts Meta at odds with the EU's vision and possibly other tech giants, highlighting the ongoing tension between innovation and regulation in the rapidly evolving AI landscape.
In a significant move that has sent ripples through the tech industry, Meta has officially announced its refusal to sign the European Union’s code of practice for its AI Act. This decision comes just weeks before the bloc’s rules for providers of general-purpose AI models are set to take effect in August 2025. Meta’s stance marks a notable challenge to the EU’s regulatory approach to artificial intelligence and raises important questions about the future of AI governance globally.
Meta’s chief global affairs officer Joel Kaplan didn’t mince words when explaining the company’s position. “We have carefully reviewed the European Commission’s Code of Practice for general-purpose AI (GPAI) models and Meta won’t be signing it,” Kaplan wrote in a statement that quickly garnered attention across the tech sector. He elaborated that after careful review, Meta determined it couldn’t sign on due to what it perceives as significant flaws in the framework. According to Kaplan, “This Code introduces a number of legal uncertainties for model developers, as well as measures which go far beyond the scope of the AI Act.”
The EU’s code of practice, published earlier in July, represents a voluntary framework designed to help companies implement processes and systems to comply with the bloc’s ambitious AI legislation. Among its key provisions, the code requires companies to provide and regularly update documentation about their AI tools and services. Perhaps most controversially for content creators and technology companies alike, it prohibits developers from training AI on pirated content and mandates compliance with content owners’ requests to exclude their works from AI training datasets.
These requirements touch on one of the most contentious aspects of modern AI development: the use of copyrighted materials for training large language models and other AI systems. Content creators across various industries have expressed concerns about their work being used without proper compensation or permission, while tech companies have generally argued for broader access to training data to advance AI capabilities.
The European Commission designed the code as a bridge during the critical interim period between when General Purpose AI model provider obligations come into effect (August 2025) and the adoption of harmonized European standards (expected by August 2027 or later). It serves as a practical guide for implementing the more abstract legal requirements of the AI Act, which takes a risk-based approach to regulating artificial intelligence applications.
Meta’s refusal to sign the code reflects deeper concerns about the EU’s regulatory approach. Kaplan characterized the implementation as “overreach” that will “throttle the development and deployment of frontier AI models in Europe.” He further claimed it would stunt European companies looking to build businesses on top of them, suggesting that Meta views the regulations as potentially harmful to innovation within the European market.
The AI Act itself categorizes AI systems based on their risk level, with outright bans on uses deemed to present “unacceptable risk,” such as cognitive behavioral manipulation or social scoring. It also defines “high-risk” applications, including biometric identification, facial recognition, and AI systems used in education and employment. For these high-risk categories, developers must meet stringent risk and quality management obligations and register their systems with regulatory authorities.
While signing the code of practice is voluntary, there are significant advantages to doing so. Companies that sign on gain additional legal protection against future accusations of breaching the AI Act. The European Commission’s spokesperson for digital matters noted that AI providers who decline to sign “will have to demonstrate other means of compliance” and consequently “may be exposed to more regulatory scrutiny.”
The stakes are particularly high given the penalties for violating the AI Act. The European Commission can impose fines of up to seven percent of a company’s annual global turnover for general violations, with a somewhat lower three percent cap for companies developing advanced AI models. For a company of Meta’s size, with billions in annual revenue, such penalties could translate to staggering amounts—potentially up to €35 million or 7% of annual worldwide turnover, whichever is higher.
Meta’s decision places it at odds with the EU’s regulatory vision and potentially with other major tech companies that might choose to sign the code. This refusal is not Meta’s first clash with European regulators. The company has previously characterized the AI Act as “unpredictable” and claimed it “goes too far” in restricting innovation. Earlier this year, Meta’s public policy representatives warned that the regulations could result in products being delayed or watered down, ultimately causing European consumers to suffer.
The timing of Meta’s announcement is particularly noteworthy given the broader political context. Tech companies from across the world, including those at the forefront of the AI race like Alphabet, Meta, Microsoft, and Mistral AI, have been fighting various aspects of these rules, even urging the European Commission to delay implementation.
For the European Union, Meta’s refusal represents a challenge to its ambition to establish itself as the global leader in responsible AI regulation. The EU has consistently positioned its AI Act as the world’s first comprehensive AI law, designed to protect citizens while fostering innovation. However, if major players like Meta refuse to cooperate with voluntary frameworks, it raises questions about the practical enforceability of the regulations.
The situation also highlights the fundamental tension between different approaches to AI governance. The EU has taken a precautionary stance, seeking to establish guardrails before AI technologies become even more deeply embedded in society. In contrast, companies like Meta appear to favor a more permissive approach that prioritizes rapid innovation and development, with regulations following later as specific problems emerge.
As August 2025 approaches, the tech industry and policy experts will be watching closely to see whether other major AI developers follow Meta’s lead or choose to work within the EU’s framework. While reports indicate that Microsoft is likely to sign the code, creating a division among tech giants, the outcome of this standoff could significantly influence not only the future of AI regulation in Europe but also how other regions approach the governance of these increasingly powerful technologies.
What remains clear is that as AI continues its rapid evolution, the question of how to effectively regulate these transformative technologies while encouraging innovation will remain at the forefront of global technology policy debates. Meta’s decision marks just one chapter in what promises to be a long and complex conversation about the future relationship between artificial intelligence, private enterprise, and public governance.
