EU Unveils AI Code of Practice with €35 Million Fines for Non-Compliance

Brussels, Belgium – The European Union has released a comprehensive Code of Practice for General Purpose Artificial Intelligence (GPAI), designed to guide businesses in complying with the landmark EU AI Act. Published on July 10, 2025, the voluntary framework aims to streamline adherence to stringent rules on transparency, copyright, and safety before the AI Act's full legal provisions take effect, with non-compliance potentially triggering fines up to €35 million or 7% of global turnover.

The Code of Practice, developed through a multi-stakeholder process facilitated by the European AI Office, serves as a practical manual for AI developers. It mandates detailed documentation of model goals, training data sources, and explanations of failure cases. This initiative seeks to transform traditionally closed AI systems into auditable and measurable entities for regulators and independent testers.

A key focus of the new guidelines is intellectual property protection. Model builders are required to implement filters to flag or remove protected text, images, or music from training data. Additionally, they must provide creators with tools to verify if their work has been used, thereby helping downstream users avoid accidental copyright infringement.

Safety and security are given equal weight within the Code. Companies must conduct "red teaming" exercises to identify and mitigate harmful outputs and establish clear processes for rolling out patches post-release. For models handling sensitive data like code or biometrics, the checklist tightens, requiring robust incident response plans and secure update channels.

While adherence to the Code of Practice is voluntary, it offers a "presumption of conformity" with the AI Act, potentially reducing administrative burdens for signatories. Providers who do not sign up will need to prove compliance through alternative, potentially more costly, methods. The EU AI Office, established to enforce the Act, will play a central role in assessing compliance and fostering trust in AI.

The AI Act's requirements for general-purpose AI models officially begin on August 2, 2025. Although the EU AI Office is expected to allow approximately one year before imposing serious crackdowns, the significant penalties for non-compliance underscore the urgency for businesses to adapt. As stated in the tweet, "non-compliance can trigger fines of up to 35M euros or 7% of global turnover."

The new regulations have drawn mixed reactions from the industry. While the EU emphasizes the importance of safe and transparent AI, some major players have voiced concerns over the perceived heavy paperwork. Notably, "40 European giants asked for a 2-year delay," but Brussels has firmly rejected these requests, signaling its commitment to the established timeline.