Anthropic updates Claude's election safeguards for 2026 US midterms

In preparation for the 2026 US midterm elections, Anthropic has revised its election safeguards for the AI model, Claude. Designed to maintain neutrality, Claude is trained to regard all political perspectives equally, supported by system prompts. Prior assessments indicated that Opus 4.7 and Sonnet 4.6 achieved scores of 95% and 96% in political engagement evaluations. The company strictly prohibits the use of Claude for misleading political campaigns and has established a Usage Policy to combat voter fraud and misinformation. Testing with 600 prompts revealed that Claude provided suitable responses 100% and 99.8% of the time. In multi-turn conversations, Sonnet 4.6 and Opus 4.7 were appropriate 90% and 94% of the time. Claude also directs users to TurboVote and initiates web searches for election inquiries 92% and 95% of the time. Anthropic works with various organizations to shape model behaviors related to free expression.