Generative AI code contradicts copyright law, industry says

A diverse group of 15 different European rightsholder organisations has warned that the General-purpose AI Code of Practice (CoP) under the EU’s AI rulebook contradicts copyright law, in a letter sent to European Commissioner for tech sovereignty, security and democracy Henna Virkkunen on Monday. 

The group includes News Media Europe, the Federation of European Publishers and the European Publishers Council. 

In September the European Commission tasked a group of independent experts to draft by April a Code of Practice on General Purpose Artificial Intelligence, which includes language models such as ChatGPT and Google Gemini. 

The code is intended to help companies comply with the AI Act's rules, including on transparency and copyright-related rules, systemic risk taxonomy, risk assessment, and mitigation measures. 

Despite defining the copyright policies that providers of general-purpose AI should implement, the draft text “introduces elements that undercut EU copyright rules”, the letter says.

“For instance, rather than establishing that providers must have lawful access to copyright content they use for training, [...] the draft suggests that AI providers may merely need to make 'reasonable and proportionate efforts' to ensure they have lawful access.”

News Media Europe says in a separate statement that the CoP should be “urgently reviewed to avoid its restrictive and selective interpretation of Copyright rules on rights reservations for text-and-data mining”. 

“As a result, the CoP risks creating backdoor legislation and rights holders are still unable to verify if and how their works are used by [General Purpose Artificial Intelligence]. Considering the draft transparency template of the AI Office, the draft measures remain largely insufficient and even counterproductive to fulfil their stated purpose, as defined in the relevant articles and recitals of the AI Act,” News Media Europe said.

The rules will have to apply to many sectors, from music to text, and therefore need to be abstract, officials involved in the drafting process told Euronews.

A third draft is set to be published mid-February. The Commission can decide to formally approve the Code via an Implementing Act under the AI Act.

The AI Act will fully enter into force in August of this year, but provisions on banned systems including facial recognition systems will start to apply as of 2 February.

In the US, copyright issues and AI models have already led to legal complaints. Comedian and author Sarah Silverman in 2023 filed a lawsuit against Meta and OpenAI, the parent company of ChatGPT, alleging that the companies illegally used her memoir to train their AI language models.