Anthropic details Claude's midterm election safeguards

Anthropic has published an update on its election safeguards for the 2026 US midterms. It covers how people use Claude for politics, new voting-information features, and independent reviews of political bias.

The company used its Anthropic Insights tool to analyze 250,000 conversations from a one-week period. In the US, about 5.75% of Claude.ai conversations touched on politics, public policy or social issues, and 0.51% were specifically about elections or voting. Requests asking Claude who or what to vote for made up about 0.01% of US conversations. Requests that broke its election integrity policies made up about 0.002%.

Later this month, Claude.ai will pull official voting information from Democracy Works into its answers for US users. Questions about polling places or ID requirements will get state-specific answers. A TurboVote banner will still appear for registration and polling queries. Claude will also show up-to-date official results on election night and in the days that follow.

Anthropic says Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 score 99% on its internal political even-handedness evaluation. Three outside groups also reviewed Claude: The Future of Free Speech, the Foundation for American Innovation and the Collective Intelligence Project. The CIP drew on a survey of more than 2,000 people in the US, UK and India.

The audits found gaps. Claude sometimes added unsolicited caveats or opposing views, for example in an advocacy email. FAI found that Claude Sonnet 4.5 was more likely to soften arguments for right-leaning positions. Anthropic says it has improved how Claude handles yes-or-no questions and short-answer requests, and it is expanding its bias evaluations to cover candidates and voting questions.

The post then turns to the Usage Policy's ban on deceiving voters or disrupting elections. The available excerpt ends partway through that section.