After three months of negotiations under increasing pressure from tech giants, on 11 March 2025, the European AI Office published the third working version of the AI Act's Code of Practice. The verdict is unequivocal: issues concerning the information space have gradually been removed from what is supposed to be a self-regulation tool for AI developers, designed to demonstrate their adherence to the principles outlined in the AI Act, which came into force on 1 August 2024.
It is vital that the Code of Practice focus on safeguaring trustworthy information — yet these protections are lacking as the right to information is not even mentioned in the text. Nor does it touch on the risks unregulated development of AI presents for trustworthy information, deepfakes being one example; the proliferation of AI-generated fake news sites; or the disinformation embedded in chatbots. The protection of fundamental rights, such as dealing with systemic risks AI could have on democratic elections, have been relegated to an appendix, and their consideration is optional.
Without prospects for addressing these fundamental issues before the final version of the text, scheduled for May, RSF, which had participated in drafting the Code, announced its withdrawal from negotiations.