EU's Strategic Move to Curtail AI Model Behavior, Including Chat-GPT and Similar Entities
The European Union (EU) has unveiled a new General-Purpose AI (GPAI) Code of Practice, a voluntary guidance tool designed to help providers comply with the EU's AI Act. The AI Act, which introduces binding rules for GPAI models, will come into effect on August 2, 2025.
The code, developed through a multi-stakeholder process involving independent experts and extensive public consultation, aims to simplify the compliance process for providers of AI systems such as OpenAI's ChatGPT-4 and Google's Gemini. It focuses on practical guidelines for implementing EU security provisions for AI models.
The key provisions of the EU's AI Act relevant to the code include safety and security, transparency obligations, and copyright. Providers must assess and manage systemic risks associated with their AI models, ensure clarity about how their models work, and respect intellectual property rights.
The code offers several benefits to AI providers who choose to adopt it. By following the guidelines, providers can transparently demonstrate their commitment to the AI Act, facilitating smoother enforcement focused on monitoring adherence to the code. The code also encourages innovation, inviting providers to engage collaboratively with the EU framework without stifling development.
The code includes chapters on transparency, copyright, and security and system risks. It provides practical guidance on meeting transparency obligations, assisting providers in documenting and communicating essential model information to users and authorities. It also offers support on copyright, addressing concerns related to the use of data in training AI and the outputs generated.
Providers who sign the code can document their "good intent" and benefit from reduced administrative burden and higher legal certainty. Those who do not adopt the code will have to develop their own approach to demonstrate compliance, which may involve greater effort.
The code is seen as an important tool to help companies transition to the new European regulatory framework. It includes a form to help providers easily record technical details for improved accessibility to supervisory authorities and downstream AI developers.
However, the code of conduct needs to be approved by the EU Commission and member states before it becomes officially binding.
The EU has adopted a comprehensive AI law, with some provisions already in effect and more to come into force in early August for new AI models. The new rules for existing systems like Chat-GPT-4 will apply from next year.
[1] European Commission. (2023). General-Purpose AI Code of Practice. Retrieved from
The General-Purpose AI (GPAI) Code of Practice, a voluntary guidance tool for AI providers such as OpenAI and Google, includes practical guidelines for implementing EU security provisions for AI models. This code, aimed at simplifying compliance with the EU's AI Act, focuses on key provisions like safety and security, transparency obligations, and copyright. By following the guidelines, AI providers can demonstrate their commitment to the AI Act, encourage innovation, and potentially benefit from reduced administrative burden and higher legal certainty. Nevertheless, the code needs to be approved by the EU Commission and member states before it becomes officially binding.
