Anthropic has published a pre-midterms update on the election safeguards it has built around Claude, following up on an earlier update from this year. The post describes how the company measures election-related usage, connects users to official voting information, pressure-tests political bias, defines misuse, and hardens its enforcement systems.

To establish a baseline for how people actually use Claude in political contexts, Anthropic used its Insights tool to analyze 250,000 conversations during a one-week period (September 21-27, 2026). The analysis found that conversations touching on politics, public policy, or social issues make up about 5.75% of all US conversations on Claude.ai, while 0.51% are specifically about elections or voting. Requests asking Claude whom or what to vote for appeared in about 0.01% of conversations, and requests that violated election integrity policies were even rarer, at roughly 0.002%. The sample also showed most political requests are informational: people follow news, research candidates, draft documents, and weigh policy arguments. Anthropic expects the share of political and election-related queries to rise as Election Day approaches.

On the voter-information side, the post notes two upcoming additions. A banner linking to TurboVote, the nonpartisan registration resource operated by Democracy Works, continues to appear for questions about registration, polling places, and election dates. Later this month, Claude.ai users in the US will start receiving official voting information from Democracy Works directly in responses, for example state-specific answers about where to vote or what ID to bring. Claude will also show up-to-date official election results on election night and in the days after.

The bias section reports that Claude Opus 5.5, Sonnet 5.5, and Haiku 5.5 are the company’s most even-handed models to date, scoring 99% on its political even-handedness evaluation, while engaging more readily with political discussions and declining fewer requests. Anthropic also commissioned three independent reviews with different methodologies: the Future of Free Speech think tank at Vanderbilt University tested Claude on hundreds of contentious-topic requests; the Foundation for American Innovation compared responses to matched left- and right-leaning prompts; and the Collective Intelligence Project surveyed more than 2,000 people in the US, UK, and India to build shared criteria for evaluating sensitive-topic responses. Most responses were rated as strong and substantive. The reviews also identified shortfalls: Claude rarely refused requests but sometimes added unsolicited caveats and counterarguments, and Sonnet 4.5 hedged unevenly, more often softening arguments for right-leaning positions. Anthropic says it has improved Claude‘s handling of direct yes-or-no questions and short-answer requests, and is tuning when and how Claude adds caveats so treatment is more consistent across the political spectrum.

The misuse section recaps the updated Usage Policy, which prohibits deceiving voters or disrupting elections, including false information about candidates or voting, impersonating election officials, and suppressing turnout. The post links this to a new guide for political campaigns that maps legitimate uses (drafting press statements and fundraising emails, debate prep, turnout-data analysis, voter outreach materials) against prohibited ones (fake social media profiles posing as real people, mass-produced engagement comments, content for news sites falsely claiming independence).

To test its defenses, Anthropic partnered with the Institute for Strategic Dialogue (ISD) and the McCain Institute to pressure-test protections in the US, Israel, and Brazil, which all hold major national elections this year. The analysts found that Claude reliably refused single harmful requests but could miss a malicious campaign broken into many individually harmless-looking steps, such as audience research, news-site construction, content drafting, and distribution planning. In response, Anthropic has added classifiers targeting novel and multi-step misuse, including fake social media accounts and news sites aimed at political discussions and financially motivated election schemes; improved detection of foreign state actor abuse; strengthened cyber protections for election infrastructure and campaigns; and built localized safeguards for the three countries.

The final sections summarize what enforcement has caught. Anthropic‘s threat intelligence report describes political, state-aligned, and commercial actors weaving AI into nearly every stage of influence operations, from planning and content creation to building distribution platforms. Two cited cases: foreign-based operators of a fake news site used Claude to produce US midterm coverage favoring Russian, Iranian, and North Korean narratives, and other threat actors used Claude to fabricate news intended to attract American audiences and generate ad revenue. Anthropic says it banned the accounts involved and improved its detection for these abuse patterns. The post closes by noting that more people are using Claude for election information than before, and that Anthropic plans to publish a broader look at how Claude was used across the election cycle in the coming months.