On Friday 28 March, the various chairs of the working groups on the General-Purpose AI Code of Practice met to review progress on the next, fourth version of the code.
The third version of the code, published on 12 March this year, had been the subject of a number of criticisms from players in the sector, particularly with regard to its requirements on copyright, the reporting of complaints and the transparency of models (see EUROPE 13597/21).
These criticisms were apparently confirmed during the new consultations that have since taken place. According to experts taking part in the discussions, “the current version of the code does not meet the legal requirements for respecting copyright”.
Other criticisms have been levelled at the grievance mechanism proposed within the code, as well as at the risks to fundamental rights and safety measures in relation to the systemic risks associated with AI models.
Earlier in the day, a coalition of authors, performers and other rights holders from across the EU’s cultural and creative sectors issued a statement totally rejecting the code as it stands.
“The GPAI’s third draft Code of Practice undermines the aims of the AI Act, contravenes European law and ignores the intention of the European legislator – we cannot support it”, they said.
“Rather than providing a robust framework for compliance, it sets the bar so low as to provide no meaningful assistance for authors, performers and other rightsholders to exercise or enforce their rights”, they write. The signatories believe that their requests and feedback have been “largely ignored or dismissed” by the Code’s drafters.
“There is a sort of area of agreement in sight. We’ve also received feedback that the code is now more specific and detailed”, says an expert involved in the drafting process.
The final contributions are due by the end of the week. The next (and usually final) version of the code is due to be presented on 2 May.
See the press release: https://aeur.eu/f/g5u (Original version in French by Isalia Stieffatre)