An update on our election safeguards

Claude, an AI model developed by Anthropic, is designed to provide accurate and unbiased information during election periods, ensuring users receive comprehensive answers to questions about political parties, candidates, and voting procedures. To achieve political neutrality, the model undergoes character training and is reinforced with system prompts that emphasize balanced responses. Before each election, evaluations are conducted to measure Claude's impartiality and effectiveness in handling prompts from across the political spectrum. These evaluations have shown high compliance rates, with Opus 4.7 and Sonnet 4.6 scoring impressively. Claude is also equipped with safeguards against misuse, such as generating misinformation or participating in influence operations. Election banners and web searches are used to direct users to reliable resources like TurboVote for up-to-date information. As part of its ongoing efforts to maintain trust, Anthropic collaborates with external organizations to review model behaviors related to freedom of expression and continuously refines its policies and detection capabilities to address real-world challenges.