EU publishes AI Code of Practice: a voluntary yet strategic guide | Festina Lente - Your leading source of AI news | Turtles AI

The EU Commission has published the Code of Practice for General Purpose AI, a voluntary but binding document. Structured in three chapters—transparency, copyright, and security—it accompanies the entry into force of the AI ​​Act on August 2, 2025.

Key points:

• Voluntary nature with legal benefits for signatories
• Three chapters focusing on transparency, copyright, and security
• Targeted application for general purpose and systemic risk models
• Deadlines: mandatory from August 2, 2025, enforcement 2026–2027

The Commission, supported by a group of 13 experts and over 1,000 stakeholders, has finalized the Code to facilitate alignment with the AI ​​Act. The text includes a chapter dedicated to transparency, one on copyright compliance—including attention to scraping and robots.txt—and a more extensive chapter on security, aimed only at providers of very high-capacity models.

The goal is to offer a simplified compliance process: companies that sign up will benefit from reduced administrative burdens and greater legal certainty, while those choosing other compliance methods will still be required to demonstrate that they meet the same requirements. Among the requirements are the compilation of standardized documentation containing technical and energy details and ten years of document retention (DB).

The timetable calls for the GPAI chapter to enter into force on August 2, 2025, with gradual enforcement: one year later for new models and two years later for existing ones. Fines for violations of the AI ​​Act can reach up to 7% of global turnover.

The Code is not yet official: it must be approved by Member States and the Commission by the end of the year. Several industry players, including the European Chamber of Commerce and major companies such as Mistral, ASML, Airbus, and Mercedes-Benz, have already requested a postponement or simplification, fearing that its implementation could penalize European companies.

Academics, digital rights associations, and privacy groups have instead called for maintaining the established deadlines, so as not to compromise their effectiveness and consistency with EU principles regarding consumers and fundamental rights. The European AI market, also supported by investment plans of up to €200 billion (~20 billion for AI "gigafactories"), will have to evaluate whether to adhere to the Code or pursue alternative avenues to demonstrate compliance.

The Code therefore represents a concrete tool for translating the AI ​​Act’s requirements into implementation commitments, while still leaving broad discretion in compliance decisions: adhering to it offers advantages, but is not mandatory.

Everything will be decided in the coming weeks, between formal approval and strategic decisions by companies, in full compliance with regulatory deadlines.