The European Union has taken another step towards finalizing its AI regulations with the release of the third draft of the Code of Practice for General Purpose AI (GPAI) models. Published on Tuesday, this draft aims to provide clearer guidance for AI providers on how to comply with the EU AI Act, particularly for large-scale AI developers. With a May deadline approaching, this version is expected to be the final revision before official adoption.
A new website has also been launched to improve accessibility to the Code, and stakeholders have until March 30, 2025, to submit feedback on the latest draft.
Navigating AI Compliance: What the Code Covers
The EU AI Act follows a risk-based approach, setting stringent rules for the most powerful AI models. These rules focus on critical areas such as transparency, copyright, and risk mitigation. The Code serves as a guide to help GPAI developers meet these legal obligations and avoid hefty penalties, which could reach 3% of global annual turnover for non-compliance.
A More Streamlined Approach
Compared to previous drafts, the latest version of the Code features a more structured framework with refined commitments and measures. This streamlining follows feedback on the second draft, published in December. Additional input from working groups and AI experts is expected to shape the final guidance, with an emphasis on clarity and coherence.
The draft is divided into key sections, covering commitments for GPAI providers, transparency requirements, copyright obligations, and safety and security measures for high-risk models.
- Transparency: The Code introduces a model documentation form that GPAI developers might be required to complete. This aims to help downstream users understand how to comply with AI regulations.
- Copyright: This remains a contentious area, with provisions requiring AI developers to make “best efforts” in respecting copyright when sourcing data. However, critics argue the vague wording may allow companies to sidestep stricter compliance.
- Safety & Security: AI models with systemic risk (GPAISR) must evaluate and mitigate potential harm. The latest draft scales back some previously suggested safety measures following industry feedback.
Ongoing Copyright Concerns
One of the most debated areas of the Code relates to copyright and AI training data. The latest draft replaces earlier provisions for rapid and direct grievance handling with a simplified point of contact for rightsholders. Moreover, it states that AI companies can reject copyright complaints if they are deemed “manifestly unfounded or excessive”. This could mean that automated mass complaints from artists and copyright holders may be ignored.
U.S. Influence and Growing Pressure
The latest revision also comes amid increasing pressure from the U.S. to scale back AI regulations. The Trump administration has criticized Europe’s AI rules, warning that overregulation could stifle innovation. At the Paris AI Action Summit, U.S. Vice President JD Vance dismissed AI safety regulations as unnecessary, advocating instead for policies that maximize “AI opportunity.”
This criticism appears to be influencing EU lawmakers, with some AI-related initiatives, like the AI Liability Directive, being scrapped. Additionally, the EU has introduced a package of reforms aimed at reducing bureaucratic hurdles, including streamlining sustainability reporting requirements.
Challenges for AI Companies
European AI firms, including French startup Mistral, have raised concerns about compliance challenges under the AI Act. Mistral co-founder Arthur Mensch recently stated that his company is struggling to meet certain technical requirements and is working with regulators to resolve these issues.
Meanwhile, the EU AI Office is preparing its own set of clarifying guidelines that will define GPAIs and their obligations more precisely. These forthcoming guidelines could potentially ease some of the regulatory burdens in response to industry and U.S. lobbying efforts.
As the final version of the Code of Practice takes shape, further guidance from the AI Office is expected “in due time” to provide additional clarity. The coming months will determine whether the EU stays firm on AI regulation or bows to industry and international pressure to relax its stance.
With AI governance at a crossroads, all eyes are on the EU as it moves to finalize one of the most comprehensive AI regulatory frameworks to date.
Read more AI news on Techvop.
