Earlier this year, Anthropic published an update on its election safeguards. That update covered how Anthropic measures and prevents political bias, detects misuse, and points people to accurate voting information. This article describes how Anthropic has built on that work ahead of the US midterms.
How people use Claude for elections and politics
Anthropic wants Claude to be a useful tool for political conversations while preventing election-related abuse. That starts with understanding how people actually use Claude in these contexts. Using Anthropic Insights, Anthropic analyzed a random sample of 250,000 US Claude.ai consumer conversations from a one-week period (September 21 to 27, 2026) in a privacy-preserving way.
Conversations touching broadly on politics, public policy, or social issues make up about 5.75% of US conversations on Claude.ai, Anthropic's consumer app. People use Claude to follow the news, research public policy, learn about candidates, and navigate government services. About 0.51% of conversations are specifically about elections or voting. Anthropic expects these shares to rise as Election Day approaches.
People rarely ask Claude whom or what to vote for. Such requests appeared in about 0.01% of US conversations. More often, people ask questions that help them reach their own conclusions, such as where candidates stand on an issue. Requests that violated Anthropic's election integrity policies were rarer still, at about 0.002% of conversations. That figure came from a separate analysis of 250,000 Claude.ai conversations reviewed against the policy.
Most requests are informational. People use Claude to make sense of geopolitical events, research candidates, draft documents, help with coursework, and weigh arguments for and against policies.
Connecting people to trusted sources
Anthropic has added two ways for people to get reliable information directly in conversations with Claude.
- As announced in April, when people ask Claude.ai about voter registration, polling places, or election dates, a banner will keep linking to TurboVote, a nonpartisan resource from Democracy Works.
- Later this month, Claude will begin drawing official voting information from Democracy Works into its responses for US users on Claude.ai. Someone asking where to vote or what ID to bring will get an answer specific to their state without leaving the conversation. The answers draw on public information that Democracy Works collects from state and local election sources.
Claude will also show up-to-date official results on election night and in the following days. People who ask for current vote counts will see the latest official numbers in the conversation.
Measuring and preventing political bias
People asking Claude about politics should get accurate, evenhanded answers that favor no political viewpoint. Anthropic tests its models extensively for political bias before every launch. Claude Opus 5.5, Sonnet 5.5, and Haiku 5.5 are Anthropic's most evenhanded models to date, scoring 99% on its political evenhandedness evaluation. The latest models also engage more readily with political discussions and decline fewer requests. Anthropic continues to refine these evaluations and look for new ways to pressure-test how its models handle political topics.
As part of that effort, Anthropic asked three independent groups to review how Claude handles free expression, each using a different method:
- The Future of Free Speech, a think tank at Vanderbilt University, tested Claude on hundreds of requests on contentious topics.
- The Foundation for American Innovation compared Claude's responses to matched left-leaning and right-leaning prompts, from mainstream policy debates to more extreme scenarios.
- The Collective Intelligence Project surveyed more than 2,000 people in the US, UK, and India. In a three-stage process, participants proposed principles for handling sensitive topics, identified topics to test, and helped develop criteria for judging responses. The group then tested for agreement across demographic and cultural groups to establish shared principles.
In the first two evaluations, most responses were rated as strong and substantively engaged with the requests.
The evaluations also found shortcomings. The Future of Free Speech and the Collective Intelligence Project found that Claude rarely refused requests but sometimes added unsolicited caveats and counterarguments, such as inserting an opposing view into an advocacy email. The Foundation for American Innovation found that Claude Sonnet 4.5 didn't always hedge evenly and was more likely to soften arguments for right-leaning positions.
Anthropic is responding in several ways:
- It has improved how Claude answers direct yes-or-no questions and requests for shorter answers, where responses could be less balanced.
- It is using the findings to improve when and how Claude adds caveats or counterarguments, so requests across the political spectrum are treated more consistently.
- It continues to improve testing for uneven hedging, building on evaluations that use matched left and right prompts.
- It is working to expand its bias evaluations to better reflect real conversations and to cover candidates and voting questions as well as policy issues.
How Anthropic defines misuse
Anthropic's Usage Policy prohibits using Claude to deceive voters or disrupt elections. That covers spreading false information about candidates or how to vote, impersonating election officials, and trying to suppress turnout. Anthropic updates the policy each year to reflect evolving model capabilities and customer feedback. This year's update, announced last week, clarifies what the election rules cover and adds new examples.
Many political campaigns use Claude for legitimate everyday tasks, so Anthropic recently published a guide to help them understand where the policies apply. The guide covers uses such as drafting press statements and fundraising emails, prepping for debates, analyzing turnout data, and building voter outreach materials. It also lists prohibited uses, such as creating fake social media profiles that pose as real people, mass-producing comments to inflate engagement, or writing content for news sites and pages that falsely claim to be independent.
Testing and strengthening defenses
Anthropic deploys automated classifiers trained to detect election-related abuse. To test these safeguards, it partnered with the Institute for Strategic Dialogue and the McCain Institute, both independent research organizations, to pressure-test protections in the US, Israel, and Brazil. Analysts tested whether Claude would help with deceptive campaigning, misinformation, and other election abuse. They began with direct requests, then broke full campaigns into smaller tasks that looked harmless alone, such as audience research, building news sites, drafting content, and planning distribution. They found that Claude reliably refused a single harmful request but could miss a malicious campaign spread across many individually harmless-looking requests.
Based on these findings and its own, Anthropic has:
- Added new classifiers to better identify novel and multi-step misuse, including attempts to build fake social media accounts and news sites aimed at political discussions, or to run financially motivated election schemes.
- Enhanced detection of election-related misuse by foreign state actors.
- Strengthened cyber protections for election infrastructure and campaigns.
- Built targeted safeguards for localized risks identified in the US, Israel, and Brazil, which all have major national elections this year.
Anthropic's threat intelligence and enforcement teams review flagged activity and act on confirmed violations, including banning accounts. Anthropic also routinely shares examples of unintended model actions or behavior observed during its evaluations and internal use of Claude.
What Anthropic has found
Anthropic's recent threat intelligence report reflects a broader shift in election-related abuse. Political, state-aligned, and commercial actors now use AI in nearly every aspect of influence operations, from planning and content creation to building platforms that spread a message. Because these early stages often happen on Claude, Anthropic can often stop efforts before they reach the public.
Anthropic's teams have continued to investigate and prevent coordinated abuse around elections, including the US midterms. In one case, foreign-based operators of a fake news site used Claude to produce midterm coverage favoring Russian, Iranian, and North Korean narratives. Anthropic also saw threat actors try to exploit the election for money, using Claude to produce fabricated news and other content meant to attract American audiences and generate advertising revenue. Anthropic banned the accounts involved and improved its ability to detect this type of abuse.
Looking ahead
More people than ever are turning to Claude for information about this year's elections. Anthropic has a responsibility to ensure the answers are accurate and balanced and that Claude isn't used to undermine the democratic process. It will keep investing in making Claude more evenhanded, preventing and addressing misuse, and being transparent. This work matters for the midterms and extends to elections worldwide. In the coming months, Anthropic hopes to share a more comprehensive look at how people used Claude throughout the election cycle, to strengthen its election-related evaluations and safeguards.