AI Chatbots Refute Election Conspiracy Theories But Generate Illustrative Images

A recent study by the Brennan Center for Justice has highlighted a concerning duality in AI chatbot behavior regarding election integrity. Researchers found that widely used AI models, including ChatGPT, Gemini, Grok, Claude, and Perplexity, consistently refuted five common election conspiracy theories when presented with them.

Despite these refusals to validate false claims such as "Noncitizens vote in large numbers" and "Voting machines are rigged," the study noted that nearly half of the responses contained inaccuracies or flawed citations. One in three responses included factual errors, broken links, or misleading references, indicating issues with accuracy even when denying misinformation.

Compounding these findings, the same technology companies behind these chatbots also offer image and video generation tools. The study found that platforms like ChatGPT, Gemini, Grok, and Meta were willing to produce images that illustrated or supported conspiracy theories upon request. Examples included generating visuals depicting "ballot harvesting."

AI-generated media is increasingly appearing in political campaigns. Recent examples include an AI-generated video in a Texas Republican Senate primary and deepfake videos used in Georgia and Massachusetts campaigns to portray politicians negatively, sometimes with fictional elements like portraying a governor as a vampire.

These developments underscore the dual-use potential of AI in the context of elections. While AI can serve as a tool to combat disinformation, its generative capabilities also pose a risk of amplifying it, presenting new challenges for maintaining an informed electorate and trustworthy information sources.