**
As the world grapples with the increasing influence of artificial intelligence in everyday life, a recent study has unveiled alarming shortcomings in the political guidance offered by AI chatbots during Hungary’s parliamentary elections. The findings, published by the civil liberties organisation Liberties, highlight a concerning trend of inaccuracies and inconsistencies that could mislead voters at critical moments.
AI’s Role in Political Decision-Making
The research, which focused on the elections held earlier this year, underscores the pitfalls of relying on AI for electoral advice. The study found that chatbots such as ChatGPT and Gemini regularly provided voters with incorrect recommendations, often suggesting parties that were not even on the ballot or failing to mention those that were. This raises significant questions about the reliability of general-purpose AI systems in contexts where informed decision-making is paramount.
The investigation revealed that when users presented detailed profiles aligned with the Tisza party—an opposition group led by Péter Magyar—AI responses frequently overlooked this major contender. In a staggering 90% of cases, the chatbots failed to recommend Tisza, highlighting a pronounced bias that could distort electoral choices. Magyar’s party ultimately triumphed, ending Viktor Orbán’s long-standing dominance, but the potential ramifications of flawed AI guidance in a more competitive environment are troubling.
Inconsistent Recommendations and Opaque Processes
The research process involved creating five distinct voter profiles based on the official party positions outlined on Voksmonitor, a well-regarded Hungarian voting advice platform. Each profile was tested multiple times across both ChatGPT and Gemini. The results were telling: Tisza garnered a mere 2% recognition rate in percentage-matching tests, while Fidesz, the ruling party, was often recommended to users aligned with its views in approximately half of the cases.
The researchers pointed out that the chatbots, despite their apparent confidence in responses, frequently delivered wildly inconsistent advice. Users received starkly different recommendations when re-posing identical queries, further eroding trust in the technology. Moreover, a staggering 96% of the responses included parties that were not part of the 2026 ballot, further muddling the electoral landscape for unsuspecting voters.
The Need for Regulatory Safeguards
The report advocates for greater regulatory oversight over AI systems that provide political advice. As it stands, general-purpose AI models, which do not disclose their decision-making processes, pose a unique risk in electoral settings. The EU’s AI Act aims to address systemic risks, while the Digital Services Act covers risks to electoral processes, yet chatbots currently slip through the cracks of these regulations.
Eva Simon, head of Liberties’ tech and rights programme, stressed the urgency for AI providers to develop robust safeguards. “General-purpose AIs should not present opaque and unstable political matching as if it were reliable electoral guidance for voters,” she stated. The need for transparency, accuracy, and consistency in AI-generated recommendations has never been more critical, especially as democracy increasingly intersects with advanced technology.
Why it Matters
The implications of this study transcend Hungary’s borders, highlighting a global concern about the role of AI in democratic processes. As voters rely more on technology for guidance, the potential for misinformation and misrepresentation looms large. Without stringent regulations and accountability mechanisms, opaque AI systems could inadvertently undermine the very foundations of democratic choice, leading to an electorate that is misinformed and misled. As we continue to integrate AI into our decision-making processes, ensuring its reliability and integrity becomes not just a technological challenge but a fundamental requirement for the health of democracy itself.
