4 AI Chatbots Fail 90% of Election News Tests in Forum AI Study
Updated
Updated · Bloomberg · May 20
4 AI Chatbots Fail 90% of Election News Tests in Forum AI Study
2 articles · Updated · Bloomberg · May 20
More than 3,100 test questions posed to ChatGPT, Gemini, Claude and Grok found their election-related answers failed on accuracy, bias or source selection 90% of the time.
Forum AI said the problems extended beyond elections to geopolitics, with the four major chatbots struggling to answer sensitive news questions fairly and accurately.
The findings come ahead of midterm elections, highlighting persistent limits in AI systems' handling of political news, healthcare and foreign-affairs topics.
AI is accurate on vaccines but fails on elections. Is the problem the technology or the nature of political truth itself?
As AI chatbots become friendlier to boost engagement, are they becoming inherently more biased and dangerously inaccurate?
With AI fueling voter manipulation and mental health crises, can society build safeguards for democracy and individual well-being simultaneously?
The AI Threat to Election Integrity: How Inaccurate Chatbots and Weak Regulation Endanger Democracy (2024–2026)
Overview
Between January and March 2024, studies revealed that AI-generated election information, especially from chatbots, was often inaccurate or misleading. Experts and election officials raised serious concerns after a study by Proof News and Princeton University found that five major AI models gave unreliable answers to questions about the 2024 U.S. election. The findings showed that AI could produce information that seemed credible but was actually false, leaving officials shocked and worried about the risks for upcoming elections. This highlights the urgent need to address the reliability and impact of AI in providing election information.