Below is this week's tracker of the latest legal and regulatory developments in the United States and in the EU. Sign up here to ensure you do not miss an update.

AI Intellectual Property Update:

  • The Copyright Review Board, for the fourth time, has refused to register an AI-generated work. The work, titled "SURYAST," (reproduced below) was created by taking an original photograph inputting that photograph into an AI painting app called RAGHAV, and then inputting a copy of Vincent van Gogh's The Starry Night as the "style" to be applied to the photograph. The Board found that "the Work does not contain sufficient human authorship necessary to sustain a claim to copyright."
    • The Board further explained that, "it is the AI model, not its user, that predicts stylizations for paintings and textures never previously observed, and that predictive function is tied to the proximity of the style image to styles trained on by the model."
    • And the mere act of selecting a style was not sufficient: "these choices only constitute an unprotectable idea for the Work, that is: an altered version of his photograph in the style of The Starry Night . . . copyright does not protect the concept reflected in a work—protection is given only to the expression of the idea—not the idea itself."
  • OpenAI and German publisher Axel Springer (owner of publications including Politico and Business Insider and European publications Bild and Welt) announced a partnership that will provide Chat GPT with summaries of news content from Axel Springer's publications, including otherwise paid content. "ChatGPT's answers to user queries will include attribution and links to the full articles for transparency and further information."

AI Litigation Update:

  • Cases brought against Meta by Sarah Silverman and Michael Chabon, among others, have now been consolidated (Kadrey v. Meta Platforms, Inc., N.D. Cal. 3:23-cv-03417). An amended complaint filed in the consolidated case has dropped claims about various infringement (from the model's outputs) as well as DMCA claims, and now only alleges direct copyright infringement relating to training the Llama model using the Books3 database. This reflects a trend in the AI copyright cases towards a focus on the model inputs and away from model outputs.
    • The complaint alleges that "a key reason Meta chose not to share the training dataset for Llama 2 was to avoid litigation from using copyrighted materials for training that Meta had previously determined to be legally problematic."

AI Policy Update—Federal:

AI Policy Update—European Union:

  • On December 8, 2023 the European Parliament and the Council of the EU reached a provisional political agreement on the proposed EU AI Act. The proposed EU AI Act has not been formally approved. The agreement reached by the European Parliament and the Council of the EU now needs to be translated into law, before being subject to the co-legislators' vote which should likely take place prior the European elections in spring 2024. The European Parliament, the Council of the EU and the European Commission all issued separate press releases, but no further information than what is contained in the press releases has been confirmed by official sources. The main elements of the agreement are as follows:
    • A definition of an "AI system" to provide sufficiently clear criteria for distinguishing AI from simpler software systems, aligning the definition with the approach proposed by the OECD;
    • Endorsement of the risk-based approach initially proposed by the European Commission;
    • Specific rules for general purpose AI models and foundation models, in particular to ensure transparency along the value chain and to address potential systemic risks;
    • Fundamental rights impact assessment for high-risk AI systems
    • Law enforcement exceptions;
    • Supervision of the implementation of the rules by national authorities, and creation of a new European AI Office within the European Commission to ensure coordination at European EU level;
    • Fines ranging from 7.5 million or 1.5% of the offending company's global turnover to € 35 million or 7% of global turnover.
  • The Court of Justice of the European Union (CJEU) ruled that "credit-scoring" amounts to "automated individual decision-making", which is, in principle, prohibited, under Article 22 of the General Data Protection Regulation (GDPR), unless certain conditions are met. In this context, the CJEU stipulated that keeping information relating to the granting of a discharge from remaining debts for longer than the public insolvency register, is contrary to the GDPR.
  • According to a letter (in Dutch) communicated to Dutch lawmakers, the Dutch government has decided that when its services want to use generative AI, there should be a prior case-by-case risk analysis, consisting of a Data Protection Impact Assessment and an Algorithm Impact Assessment.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.