Experiments using models like GPT-4o and DeepSeek found Trump supporters moved toward Kamala Harris by nearly four points on a 100-point scale before the 2024 U.S. election.

Humans may view other humans as fallible while trusting machines more, complicating defenses against AI persuasion and underscoring the need for AI literacy.

The findings feed into broader debates about misinformation, algorithmic influence, and electoral integrity without claiming guaranteed outcomes.

The articles advocate guardrails, including auditing and documenting the accuracy of LLM outputs in political conversations to protect democratic processes.

Scholars emphasize the need for regulation and safeguards as AI-mediated persuasion grows, calling for ongoing scrutiny.

Researchers stress robust safeguards and further study as AI tools become more prevalent in shaping opinions and decisions.

The studies cite works by Lin et al. in Nature and Hackenburg et al. in Science, noting authors’ lack of competing interests.

Limitations to manipulation include users’ time constraints and hard psychological limits to persuadability, suggesting lab results may overstate real-world impact.

A pair of studies published in Nature and Science show AI chatbots can influence political opinions across the U.S., Canada, and Poland by engaging people in conversations that advocate for a candidate.

Researchers note AI tools can reduce susceptibility to conspiracy theories when they present reasoned arguments, but risk emerges as models fabricate content if information runs dry.

Independent experts caution about real-world applicability and risks, including misuse to spread radical ideologies, while acknowledging potential legitimate uses with transparency.

Experts warn these AI persuasion tools could become a central democratic concern in elections and call for safeguards to prevent manipulation.