Proposed EU AI Act—Presidency compromise text: select overview and comment on the changes to the proposed regulation

Abstract

With its proposed EU AI Act, the EU is aspiring to lead the world in admiral AI regulation (April 2021). In this brief, we summarise and comment on the ‘Presidency compromise text’, which is a revised version of the proposed act reflecting the consultation and deliberation by member states and actors (November 2021). The compromise text echoes the sentiment of the original text, much of which remains largely unchanged. However, there are important shifts and some significant changes. Our main comments focus on exemptions to the act with respect to national security; changes that seek to further protect research, development and innovation; and the attempt to clarify the draft legislation’s stance on algorithmic manipulation. Our target readership for this paper is those who are interested in tracking the evolution of the proposed EU AI act, such as policy-makers and those in the legal profession.

Typology of Risks of Generative Text-to-Image Models

“…These tools could potentially address some socio-legal concerns associated with TTI systems and similar generative model-induced technologies, including data privacy, intellectual property infringement, and security vulnerabilities [70,138,161]. For instance, the EU AI Act can help provide a legal framework for the responsible use of TTI systems, setting out the rights and responsibilities of different stakeholders [53,73,87,101]. Privacy laws might be adjusted to regulate the collection, storage, and use of personal data used to train or operate TTI models, thereby safeguarding individual privacy Samuelson [138].…”

“…There are a number of regulatory efforts currently addressing data access and the use of AI, with modifications underway to incorporate generative technologies like TTI models. These include the EU AI Act [53,73,87,101], the Algorithmic Accountability Act in the US [108], and China's Deep Synthesis Provisions [80], among others. Multiple ongoing lawsuits could shape future legal perspectives on generative models, including TTI-induced systems.…”

Mitigating the adverse effects of AI with the European Union's artificial intelligence act: Hype or hope?

“…13 Smuha et al (2021) also criticize that the proposal operationalizes human rights in an excessively technocratic-paternalistic manner, that it overlooks the distinct nature of those rights, fails to give adequate expression to the spirit and purpose of fundamental rights and that the risk categorization remains 'blunt' and 'simplistic.' 14 The current definition of AI is too loose and can still be read as allencompassing; e.g., it is unclear how AI systems differ from traditional software (Kazim et al, 2022). Furthermore, it is unclear when an AI system becomes a general-purpose system and what the difference is (if there is any) between generative AI and general-purpose AI.…”

“…For more information on the Presidency Compromise Text published in late 2021, seeKazim et al, 2022. …”

Tecnologia disruptiva e segurança pública: uma análise da produção científica mundial

“…O relatório Proposed EU AI act -Presidency compromise text: select overview and comment on the changes to the proposed regulation (Kazim et al, 2022) resume e comenta o "Presidency compromise text" (Union, 2021), uma versão revisada da proposta de lei que reflete a consulta e deliberação pelos estados membros e atores (novembro de 2021). Com a proposta de Lei de Inteligência Artificial da União Européia existe a aspiração dessa em liderar o mundo na regulamentação da AI (abril de 2021).…”