Event Recap: AI, Foreign Interference, and the Future of Election Information

PCFR

1 day ago

3 min read

October 6, 2026 | Virtual Program

As artificial intelligence transforms how Americans access and consume information, new vulnerabilities are emerging that could have significant implications for democratic institutions. On October 6, the Phoenix Committee on Foreign Relations (PCFR) hosted a virtual briefing on these topics featuring David Salvo, Managing Director for Transatlantic Policy and Programming, and Max Read, Director of Civic Innovation, at the Institute for Strategic Dialogue (ISD-US). Moderated by Caroline Lynch, Chair of PCFR's Board of Directors, the discussion explored the evolving tactics of foreign influence operations and presented new research into the accuracy and reliability of AI-powered chatbots as sources of election information.

AI and the Evolution of Foreign Influence Operations

Foreign interference in American elections is not new, but artificial intelligence is fundamentally changing the scale, sophistication, and accessibility of influence operations. The discussion highlighted how Russia, China, and Iran are using AI-enabled tools to advance distinct geopolitical objectives and influence American public discourse. Rather than focusing exclusively on national political debates, foreign actors are increasingly targeting state and local communities, exploiting local grievances and information gaps to make their messaging more credible and relevant.

Can Americans Trust AI Chatbots for Election Information?

A central focus of the briefing was ISD's research into the reliability of major AI chatbots in providing accurate information about U.S. elections.

Drawing on findings from the report Chatbots in the Ballot Box, the discussion examined a study testing six leading AI chatbots across ten states, including Arizona. The research evaluated whether these systems could accurately answer practical questions about voting procedures, registration requirements, and election administration.

The findings revealed significant shortcomings:

Almost 30% of chatbot responses to English-language prompts on election questions were incomplete, inaccurate, or otherwise problematic.
Of these, 12% of responses contained outdated or misleading information serious enough to potentially prevent voters from casting valid ballots.
Spanish-language responses were, on average, 16% less accurate than English-language responses.

The findings demonstrated how seemingly routine questions fed to LLMs about registration deadlines, voter eligibility, or ballot procedures could generate misleading answers with real-world consequences.

Emerging Vulnerabilities and the Threat of Manipulation

Beyond inaccuracies generated by AI systems themselves, the discussion explored the possibility that foreign actors could deliberately manipulate the information chatbots provide.

One particular vulnerability involves so-called "data voids"—areas where reliable, authoritative information is limited or difficult for AI systems to retrieve. Malicious actors may seek to fill these gaps with fabricated or misleading content, potentially influencing the answers generated by AI-powered tools.

Looking Ahead: Protecting Election Information

The conversation concluded by examining the need for continued research, stronger safeguards, and greater public awareness as AI systems become increasingly influential sources of information. The speakers provided an excellent set of readings based on their research at ISD for further perusal by PCFR members.