Claude to show official voting data to US users

Claude will start this month to use official voting information from Democracy Works, a nonpartisan organization, when a user in the United States asks about the November midterm elections. Anthropic, which develops Claude, announced it in a post published Oct. 14. The company gave no exact date and did not say whether the feature will reach other countries.

With that information, anyone asking where to vote or which ID to bring will get an answer tailored to their state. The company adds that on election night and in the days after, Claude will present up-to-date official results. For questions about registration, polling places or deadlines, the assistant already showed a notice linking to TurboVote, a resource from the same organization.

What the company measures

Anthropic reviewed a sample of about 250 000 conversations with US users between Sept. 21 and 27, using a method it says preserves privacy. These are the company's own figures, with no independent verification:

• About 5.75% touched on politics, public policy or social issues.
• About 0.51% touched on elections or voting.
• About 0.01% asked whom to vote for.

A separate review of another 250 000 conversations found about 0.002% that violated its election integrity policies. Anthropic also says its three current models (Opus 5.5, Sonnet 5.5 and Haiku 5.5) score 99% on its internal political even-handedness evaluation.

Measures against abuse

The post describes classifiers that detect election abuse, including campaigns split into tasks that look harmless, and it bans deceiving voters or impersonating election officials. Tests by the Institute for Strategic Dialogue and the McCain Institute in the US, Israel and Brazil showed Claude reliably refused isolated harmful requests but could miss those fragmented campaigns. The company says it blocked accounts from a fake news site run outside the US that used Claude to cover the elections.

For users in Mexico, nothing changes today. Still pending are the activation date and the fixes Anthropic says it is preparing after criticism from outside reviewers, who found unrequested caveats and uneven treatment of right-leaning arguments in an earlier version of the model.