Anthropic says it has added detection tools for election abuse ahead of the US midterms. Its October 14 update also promises official voting information inside Claude.ai later this month, following last week’s usage policy announcement.
Voting information and campaign rules
Democracy Works will supply state-specific voting information for US users. Anthropic also plans to show current official election results in conversations.
Campaigns may use Claude for fundraising emails, debate preparation and turnout analysis. Prohibited uses include deceiving voters, impersonating election officials, fake social profiles and news pages that falsely claim independence.
Ordinary tasks can conceal coordinated abuse
According to Anthropic, testers at the Institute for Strategic Dialogue and McCain Institute found Claude reliably refused direct harmful requests. It could miss malicious campaigns split into tasks such as research, website building and content drafting. Anthropic says it added classifiers to detect multi-step misuse.
Election or voting queries accounted for about 0.51% of 250,000 US Claude.ai consumer conversations sampled September 21 to 27. A separate analysis of 250,000 Claude.ai conversations found election-policy violations in about 0.002%.
Sources
- Anthropic: Preparing for the 2026 US midterms. Lead chart by Anthropic, from the same page.
