The recently published EU Code of Practice on General-Purpose AI (GPAI) offers little clarification or additional guidance on preventing and mitigating AI risks for electoral processes.
While we welcome the explicit reference to “democratic processes” and “harmful manipulation” in the final version of the Code as an essential starting point, the current framing remains too vague to capture the depth and complexity of risks that GPAI poses to democracy, particularly regarding the following areas:
1. Specification of democracy-related risks: the language used remains abstract and lacks actionable clarity. To ensure meaningful risk mitigation, the taxonomy should identify concrete threats that GPAI systems pose to democratic processes, particularly elections.
2. Strengthening mitigation measures: the GPAI CoP currently under-specifies mitigation measures tailored to democratic contexts.
4. Civil society marginalisation: While the EU’s stated goal for the drafting of the Code of Practice was to include various stakeholders, the actual process fell short-particularly with regard to civil society participation. Only 11% of participants represented civil society organisations, while nearly half were from industry.
5. Effective enforcement mechanisms: The fact that major actors like Meta can opt out of the Code without consequence further reinforces the need for binding legal frameworks to ensure consistent compliance and protect the public interest. In a democratic system like the EU, soft law should serve as a complement, not a substitute for enforceable regulation.
Background
The EU Code of Practice on General-Purpose AI is a voluntary framework developed by the European Commission in collaboration with industry, civil society, and other stakeholders to promote the responsible development and deployment of general-purpose AI systems. The Code aims to establish common principles and risk mitigation measures for GPAI models, particularly in areas where their use may pose systemic societal risks, including to democratic processes. It is designed to complement the implementation of the Artificial Intelligence Act (AI Act), the EU’s new legal framework for AI, which entered into force in August 2024.
Cover photo: © Rawpixels.com on Adobe Stock.
