A recent investigation into the use of AI chatbots during Hungary’s April elections has unveiled a troubling reality: these digital assistants provided voters with inaccurate and inconsistent political recommendations. The study, conducted by the civil liberties group Liberties, raises significant concerns about the reliability of general-purpose AI systems in electoral contexts, especially as their authority continues to grow.
AI’s Electoral Performance Under Scrutiny
The research highlighted how AI chatbots, particularly ChatGPT and Gemini, failed to offer sound electoral advice to users seeking guidance on which political party to support. In a striking 90% of instances involving Tisza, the opposition party that ultimately triumphed in the elections led by Péter Magyar, the chatbots did not recommend it at all. Instead, users often received suggestions for lesser-known parties that were unlikely to secure parliamentary representation or were not on the ballot at all.
This discrepancy is alarming, especially considering that 29.8% of Hungary’s population engages with AI technologies. The study tested five distinct voter profiles aligned with the main political parties, using a respected voting advice app, Voksmonitor, to create prompts for the chatbots. Despite the clarity of the profiles, the results were anything but straightforward—especially for Tisza supporters.
Inconsistent Guidance and Opaque Responses
The findings revealed a worrying pattern: the AI systems provided radically different party recommendations for identical voter profiles in successive tests. In percentage matching tests, Tisza was only recognised in a mere 2% of cases, while Fidesz, the ruling party, was frequently identified as the primary option. In fact, ChatGPT suggested Fidesz as the party to vote for in approximately 50% of cases, further illustrating a pronounced bias in the AI’s recommendations.
Moreover, 96% of the responses included parties not even running in the 2026 elections, indicating a significant gap in the chatbots’ training data and understanding of the current political landscape. Despite these inconsistencies, both AI models often prefaced their advice with disclaimers stating they could not provide political guidance, only to follow up with detailed recommendations that appeared polished and authoritative.
Regulatory Gaps in AI Oversight
The report pointed out a critical regulatory void, as current EU legislation does not adequately address the risks posed by AI chatbots in electoral processes. The EU’s AI Act mandates that AI providers evaluate systemic risks, while the Digital Services Act covers broader systemic risks affecting elections. However, AI chatbots seem to fall into a grey area where neither law fully applies.
As the findings imply, there is an urgent need for implementing safeguards for AI systems that provide political advice. Eva Simon, head of Liberties’ tech and rights programme, emphasised that such systems should not present unreliable electoral guidance to voters as if it were trustworthy.
“Democracy cannot rely on opaque systems that claim neutrality while delivering advice they cannot explain, reproduce or guarantee to be accurate,” Simon stated, highlighting the importance of transparency and accountability in AI applications.
Why it Matters
The implications of these findings extend far beyond Hungary’s borders. As AI technologies increasingly permeate the electoral landscape, the potential for misinformation and manipulation grows. With voters relying on these tools for guidance, the risk of undermining democratic processes is ever-present. Ensuring that AI systems adhere to standards of accuracy, transparency, and reliability is crucial for maintaining the integrity of elections worldwide. As we navigate this uncharted territory, it is imperative to establish robust frameworks that safeguard the democratic process against the pitfalls of technological advancement.
