Anthropic outlined new election safeguards for the 2026 US midterms. It analyzed 250,000 US conversations to understand political use. About 5.75% of US conversations touch on politics. Only 0.51% focus on elections or voting. Requests for voting recommendations are rare at 0.01%. Policy violations are even rarer at 0.002%. Claude now integrates Democracy Works data for voter registration and polling places. It will also show official election results in conversations. New classifiers detect multi-step election misuse. Foreign actors were caught using Claude for fake news. Anthropic banned those accounts and improved detection. New models score 99% on political evenhandedness tests. Independent groups identified areas for improvement in bias handling.
What changed
Claude directed users to external resources for voting information and did not integrate official data directly.
What it unlocks
Users can now get state-specific voting information and election results directly in Claude.ai conversations.
- 5.75% of US Claude.ai conversations touch on politics
- 0.51% of US conversations are about elections or voting
- 0.01% of US conversations ask whom or what to vote for
- 0.002% of conversations violated election integrity policies
What you need to act on it
- US-based user
- Claude.ai consumer app
Sources