A team of experts appointed by the EU Commission is preparing a draft Code of Practice (CoP) on General Purpose AI. This draft is scheduled for publication in April and aims to align with the EU’s AI Act. The CoP is expected to address transparency, copyright compliance, systemic risks, risk assessment, and mitigation measures for AI systems, including language models like ChatGPT and Google Gemini.
However, 15 European rightsholder organizations, including News Media Europe, the Federation of European Publishers, and the European Publishers Council, have criticized the draft. In a letter to European Commissioner Henna Virkkunen, they expressed concerns that the CoP contradicts EU copyright law.
Concerns Over Copyright Compliance in Draft Code
The draft CoP aims to define copyright policies for general-purpose AI providers. However, the letter argues that the draft undermines EU copyright law.
“For example, instead of requiring providers to secure lawful access to copyrighted content for training, the draft suggests they only need to make ‘reasonable and proportionate efforts,’” the letter stated.
News Media Europe separately criticized the CoP for its restrictive and selective interpretation of copyright rules. They warned that it risks creating loopholes and leaves rights holders unable to verify how their works are used by AI systems.
The organization also said the transparency template proposed by the AI Office in the draft remains insufficient. According to News Media Europe, these measures fail to meet the standards defined in the relevant articles of the AI Act.
Next Steps and Broader Implications
The EU Commission plans to publish the third draft of the CoP by mid-February. Officials involved in the drafting process emphasized that the rules must remain abstract to cover various sectors, including music and text.
The Commission may formally approve the CoP through an Implementing Act under the AI Act. The AI Act itself will take full effect in August, though bans on specific systems, including facial recognition, will apply from 2 February.
Meanwhile, copyright disputes involving AI systems have already sparked legal battles in the United States. For instance, Sarah Silverman filed a lawsuit in 2023 against Meta and OpenAI, alleging they used her memoir without permission to train AI models.
The ongoing debates in the EU and the US underscore the growing challenges in balancing AI innovation with copyright law enforcement.