Voters using artificial intelligence for election information may receive incomplete or biased responses tailored to their identity. Before Alaska's primary election, Claude from Anthropic gave different answers about Dan Sullivan's health care position based on whether the user was a Democrat or Republican, focusing on the incumbent senator rather than all candidates with that name. MIT researchers launched a data collection effort documenting how major AI systems respond to election queries, finding that nearly 30% of chatbot answers were incomplete, unclear, inaccurate, or outdated.

Political experts are studying how these AI systems influence democratic processes. "A. I. is becoming part of the way people encounter and make sense of political information, and yet we know relatively little about what that information environment actually looks like, how it differs for different user demographics, political identities and geographies," said Chara Podimata, co-leading the M. I.

T. project with Adam Berinsky and Charles Stewart III. The researchers are testing how nearly a dozen large language models respond to fixed questions about midterm candidates, varying responses based on user identity characteristics.