Anthropic says Claude.ai will soon answer U.S. voters’ questions about where to vote and what identification to bring with state-specific information from Democracy Works. It also plans to show official vote counts on election night and afterward. Anthropic said the voting-information integration would start later in October; neither feature was live on Oct. 14.
Democracy Works compiles public information from state and local election sources, Anthropic says, and its Elections API documentation describes a service for election information and voting guidance. Anthropic’s Oct. 14 post does not identify the provider for election-night results.
What Anthropic reports about election use and misuse
Anthropic says its analysis of a random sample of 250,000 U.S. Claude.ai consumer conversations from Sept. 21–27 found that 0.51% were specifically about elections or voting. The company classified the conversations using a privacy-preserving method; the figures are its own estimates, not an independent measurement of all voter use.
In separate tests with the Institute for Strategic Dialogue and the McCain Institute, Anthropic says Claude reliably refused direct requests for election abuse but sometimes failed to spot a campaign split into many ordinary-looking tasks. A sequence might ask for audience research, a news site, separate pieces of content and a distribution plan. The company says it has added classifiers to detect these multi-step patterns, fake accounts and sites, and financially motivated schemes.
Anthropic also reports that foreign-based operators used Claude to produce U.S. midterm coverage for a fake news site favoring Russian, Iranian and North Korean narratives. It says it banned the accounts and improved detection; its announcement does not quantify how many people saw the material.
The post names the testing partners and countries but gives no test size or outcome rates and links no detailed external evaluation.
Sources