Preparing for the 2026 US midterms
Earlier this year, we shared an update on our election safeguards, which described how we measure and prevent political bias, detect misuse, and direct people to accurate voting information.
TL;DR
- Anthropic has analyzed 250,000 US Claude.ai conversations to understand political and election-related queries, finding that direct voting recommendations are rare.
- Claude will now provide official voting information, including polling place details and ID requirements, directly within conversations via integration with Democracy Works.
- Election night results will also be directly available in Claude's responses.
- Anthropic has tested its models extensively for political bias, with recent models scoring 99% on political evenhandedness evaluations.
- Independent groups from Vanderbilt University, the Foundation for American Innovation, and the Collective Intelligence Project reviewed Claude's handling of political topics and free expression.
- The Usage Policy has been updated to clarify prohibitions against using Claude for voter deception and election disruption.
- New automated classifiers have been deployed to detect multi-step and novel forms of election-related abuse, including attempts to create fake news sites and social media accounts.
- Anthropic partnered with the Institute for Strategic Dialogue and the McCain Institute to pressure-test its election safeguards in the US, Israel, and Brazil.
- The company has observed AI being incorporated into influence operations for planning and content creation, and has banned accounts involved in using Claude for fake news and financial schemes related to elections.
- Anthropic is committed to continued investment in making Claude more evenhanded, preventing misuse, and maintaining transparency.