Photo: IMEN BEN YOUSSEF / AFP / Getty Images
Anthropic announced plans to integrate U.S. voting information into its Claude AI chatbot before the upcoming midterm elections. The company will incorporate election results and voter data from Democracy Works, a nonpartisan nonprofit organization, later this month. Anthropic aims to make Claude AI a valuable tool for political discussions and to prevent election-related misuse.
According to Anthropic, political conversations account for about 5.75% of all interactions on Claude.ai, with less than 1% specifically about elections or voting. The company expects these numbers to rise as Election Day approaches.
Anthropic's recent analysis of 250,000 conversations revealed that requests about whom or what to vote for appeared in only 0.01% of U.S. conversations. Most users asked questions that could help them form their own opinions, such as candidates' positions on issues. The company has implemented safeguards to ensure Claude provides accurate and unbiased information, drawing on data from Democracy Works to offer state-specific voting details.
Claude will also display up-to-date election results on election night and in the following days. Anthropic has tested its models extensively for political bias, achieving a 99% score on its political evenhandedness evaluation. The company continues to refine its evaluations and safeguard measures to ensure Claude remains a reliable resource for voters.
In collaboration with organizations like the Institute for Strategic Dialogue and the McCain Institute, Anthropic has enhanced its ability to detect and prevent election-related misuse, including deceptive campaigning and misinformation. The company has also strengthened its cyber protections for election infrastructure and campaigns.
Anthropic's efforts to improve Claude's political evenhandedness and prevent misuse are part of a broader initiative to ensure the chatbot supports democratic processes during the U.S. midterms and beyond.