As the risks posed by increasingly powerful AI systems grow (bias, opacity, democratic erosion), governments are under pressure to develop frameworks that can both enable innovation and safeguard public interest. In 2024 alone, 148 AI-related bills were passed globally, signalling the urgency of regulation in this field (Maslej et al. 2024, 376). Within the EU, the AI Act has emerged as a landmark attempt to legislate this balance. To translate its principles into practice, the European Commission is developing a Code of Practice for General Purpose AI (GPAI), a soft law instrument intended to guide companies in aligning with the Act. This Code has brought together a wide range of stakeholders, including academia, institutions, law makers, private companies, and civil society representatives. But while this process appears inclusive on paper, its democratic legitimacy is already under scrutiny. This essay will argue that although civil society is formally included in the drafting of the Code, its actual influence remains limited.
The absence of binding safeguards has allowed AI systems to be deployed without adequate oversight, raising concerns over privacy, discrimination and accountability. Scholars argue that regulation must not only respond to these risks but anticipate them, ensuring that innovation is shaped by ethical and democratic principles rather than market forces alone (Guio, 2020; Wirtz, Weyerer & Geyer, 2019). In this context, regulatory governance has emerged as a preferred model for structuring AI oversight (Yasuda, 2022; Leal, 2021), acknowledging the limits of state-centric governance and embracing more participatory approaches (Bareis and Katzenbach, 2022, p.856). Djefall, Siewert and Wurster (2022) distinguish between four government postures: entrepreneurial, market-oriented state, regulatory state, and the self-regulation-promoting state. The latter, relying on voluntary codes and soft law, has gained traction in AI governance, especially in multi-stakeholder initiatives where the role of the state is facilitative rather than directive. Within this framework, civil society has become increasingly important, often acting as a counterweight to dominant discourses. Sara Pane (2025) highlights their role in the “post-digital” European public sphere, fostering feedback loops between citizens and institutions and challenging the insularity of EU decision-making processes.
This essay considers the following research question: to what extent can civil society meaningfully influence the development and implementation of the Code of Practice? It will examine that influence in three stages: the democratic value of civil society participation, its role in shaping the Code, and the limits of its impact.
Why does civil society matter?
Civil society plays a crucial role in AI governance by bringing democratic legitimacy, oversight and ethical scrutiny to a process often dominated by powerful private interests. Civil society organisations (CSOs) act not only as advocates but also as watchdogs, critics, and bridge-builders between citizens and institutions. By defending human rights and amplifying public concerns, CSOs help ensure that AI policies do not drift away from democratic principles. This is particularly important in the EU where legitimacy increasingly depends on transparency and inclusiveness. Article 10(3) of the TEU calls for decision-making to occur “as closely as possible to the citizen”, reinforcing the need to involve civil society in shaping policy. Given the complexity and potential harms of AU, civil society’s contribution is essential. As Sartor (2019) argues, public enforcement alone is not sufficient in the age of AI. CSOs must expose abuses, inform public debate, and pressure institutions to act. This is a shift from traditional top-down regulation toward “regulatory governance”, in which multiple stakeholders shape, monitor and enforce rules (Yasuda, 2022; Leal, 2021). Civil society engagement is therefore not functional but normative. This is particularly relevant as decisions about AI’s design and deployment are often made by opaque corporate actors far removed from democratic accountability (Nahr et al., 2025). In short, civil society brings the public voice into a privatised policy space.
Civil society’s role in influencing the Code of Practice
The EU has formally included civil society in the drafting of the AI Code, offering a model of co-regulation that values inclusive dialogue, or at least, on paper. The process for developing the Code has been structured as a co-regulatory initiative involving more than 1000 stakeholders from diverse backgrounds. This participatory effort reflects the EU’s broader commitment to multi-stakeholder governance, echoing models like the “Quadruple Helix”, which positions civil society alongside the state, business, and academia as a core partner in innovation and policy design (Carayannis and Campbell, 2009). Through workshops, consultations, and public feedback channels, civil society has contributed to deliberations on key principles like transparency, fairness, and risk mitigation. Advocates like Moes and Friedlaender (2024) have stressed that this process is essential to produce balanced, ethical and future-proof AI regulation. Civil society’s involvement has also helped introduce issues often overlooked by industry actors, such as systemic discrimination, algorithmic opacity or the protection of marginalised communities. This diversity of perspectives is necessary not only for ethical reasons, but also for ensuring the robustness and legitimacy of the resulting Code. As the WHO (2024) and others have noted, effective governance of large-scale AI systems must reflect pluralistic values and democratic input.
Civil society’s limited impact and influence
Despite formal inclusion, civil society’s influence in the Code process is structurally constrained by power asymmetries, the non-binding nature of soft law persistent corporate lobbying. While the process appears participatory, the reality of influence suggests otherwise. CSOs have repeatedly voiced concerns that their input is being sidelined in favour of industry interests. In March 2025, the third draft of the Code triggered public backlash and a joint letter from CSOs and AI Act co-legislators. Their main critique was the downgrading of key human rights risks from compulsory to optional categories. These revisions suggest that pressure from major AI companies has effectively diluted the Code’s ambition, raising fears of regulatory capture. A central point of contention is the treatment of “systemic risk”, a legal designation in the Act for models with far-reaching impact. The Code was meant to clarify which risks developers must assess under this category. Yet in the latest draft, most risks are no longer mandatory for assessment. CSOs have described this weakening as a betrayal of the AI Act’s original spirit. This situation reveals the core weakness of the co-regulatory and soft law approach. As Arbel et al. (2023) emphasise, soft law lacks enforceability and is especially vulnerable to manipulation by the very actors it seeks to regulate. Without binding obligations, developers may ignore critical risks and civil society is left with few tools to hold them accountable. As MEP van Sparentak noted, making the protection of democracy and rights optional contradicts the very foundations of the regulation. This shift exposes the limits of civil society’s influence in a process tilted heavily in favour of industry.
The co-regulatory model behind the EU’s GPAI Code of Practice promises inclusivity, but its execution reveals the structural limits of civil society’s influence. Despite contributing to the process, civil society lacks the power to counterbalance lobbying from well-sourced tech firms. In short, civil society has been given a seat at the table but the table may already be set. For civil society to meaningfully shape AI governance, it must be empowered with real accountability mechanisms.
Bibliography
Arbel, Yonathan A., et al. “Systemic Regulation of Artificial Intelligence.” 56 Ariz. St. L. J. 545, Research Paper No. 582. AI Safety Legal Paper Series 12-24, University of Utah College of Law, 16 December 2023, https://arizonastatelawjournal.org/wp-content/uploads/2024/08/Arbel_PUB.pdf.
Carayannis, Elias G., and David F.J. Campbell. “’Mode 3′ and ‘Quadruple Helix’: toward a 21st century fractal innovation ecosystem.” International Journal of Technology Management (IJTM), vol. 46, no. 3/4, 2009.
Filgueiras, F. “The politics of AI: Democracy and authoritarianism in developing countries.” Journal of Information Technology & Politics, vol. 19, no. 4, 2022, pp. 449–464, https://doi.org/10.108 0/19331681.2021.2016543.
Friedlaender, Daniel. Europe’s AI Future: The Challenges and Opportunities of the AI Act’s Code of Practice. Interview recorded at the second edition of the European AI Roundtable on 16 October 2024, hosted by the Computer & Communications Industry Association (CCIA Europe). Computer & Communications Industry Association, 15 November 2024, https://www.youtube.com/watch?v=xgkAgA6Wv1w.
Gellers, J. C., and D. J. Gunkel. “Artificial intelligence and international human rights law: Implications for humans and technology in the 21st century and beyond.” In A. Zwitter & O. J. Gstrein (Eds.), Handbook on the politics and governance of big data and artificial intelligence (pp. 430–455), Edward Elgar Publishing, 2023, https://doi.org/10.4337/9781800887374.00027.
Kaye, D. Report on artificial intelligence technologies and implications for freedom of expression and the information environment. United Nations Human Rights Office of the High Commissioner, 2018, https://www.ohchr.org/EN/Issues/%20FreedomOpinion/Pages/ReportGA73.aspx.
Krebs, S., et al. “All the news that’s fit to fabricate: AI-generated text as a tool of media misinformation.” Journal of Experimental Political Science, vol. 9, no. 1, 2022, pp. 104-117, https://doi.org/10.1017/XPS.2020.37.
Leal, A. P. “Collaborative Regulation: Which is the role of the regulator in collaborative regulation?” Law, State and Telecommunications Review, vol. 13, no. 1, 2021, pp. 40-69.
Lee, K. F. AI superpowers: China, Silicon Valley, and the new world order. Houghton Mifflin Harcourt, 2018.
Maslej, N., et al. The AI Index 2024 Annual Report. Stanford University, AI Index Steering Committee, Institute for Human-Centered AI, Avril 2024, https://aiindex.stanford.edu/wp-content/uploads/2024/05/HAI_AI-Index-Report-2024.pdf.
Moes, Nicolas. Building Trust in AI: Civil Society’s Role in Drafting the AI Act’s Code of Practice. Interview recorded at the second edition of the European AI Roundtable on 16 October 2024, hosted by the Computer & Communications Industry Association (CCIA Europe). Computer & Communications Industry Association, 15 November 2024, https://www.youtube.com/watch?v=KZlSQ5S_-G0.
Nahr, Hendrik, et al. AI & Democracy: Exploring the democratic challenges, risks and opportunities brought by generative AI systems. Roundtable synthesis report. Make.org & ifok, 2025, https://assets.make.org/consultations/make.org_ifok_AI_and_Democracy_20250407.pdf.
Pane, Sara. “The european ‘post-digital’ public sphere: foundations of an emerging paradigm in the social sciences.” methaodos.revista de ciencias sociales, 13(1), 13(1) ed., 2025, https://doi.org/10.17502/mrcs.v13i1.866.
Sartor, Giovanni. Artificial intelligence: Challenges for EU Citizens and Consumers. Artificial intelligence: Challenges for EU Citizens and Consumers. European Parliament, January 2019, http://europarl.europa.eu/RegData/etudes/BRIE/2019/631043/IPOL_BRI(2019)631043_EN.pdf.
Settle, J. E. Frenemies: How social media polarizes America. Cambridge University Press, 2018, https://doi.org/10.1017/ 9781108560573.
World Health Organization. “Ethics and governance of artificial intelligence for health. Guidance on large multi-modal models.” 2024, https://iris.who.int/bitstream/handle/10665/375579/9789240084759-eng.pdf.Yasuda, J. “Regulatory governance.” Handbook on Theories of Governance, Edward Elgar Publishing, 2022, pp. 472-484.
