The measures include integrating official voting information into chatbot responses, providing updated election results, and improving systems to detect coordinated AI campaigns that try to influence voters.
The artificial intelligence (AI) company said it is bringing new safeguards to prevent election-related misinformation, political manipulation and attempts to interfere with voting. Anthropic is preparing Claude for the US midterm elections.
It follows investigations that revealed attempts to use Claude to produce misleading election coverage and fake news content. Anthropic analysed about 250,000 conversations over one week to understand how people use Claude for political queries. Discussions about politics, public policy and social issues accounted for 5.75% of conversations on Claude.ai in the US, while election and voting queries represented 0.51%. Requests asking Claude whom or what to vote for accounted for around 0.01% of conversations, while requests violating its election integrity policies appeared in approximately 0.002%. , according to the company.
The company said it plans to make Claude useful in political conversations while also preventing abuse that could harm the democratic process. Anthropic is also expanding its political-bias measures to ensure Claude responds more evenly across political viewpoints.
Anthropic plans to share updated election results and vote counts directly in conversations on election night and in the days that follow. Anthropic said Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 scored 99% on its political evenhandedness evaluation. The company also worked with independent organisations, including Vanderbilt University’s Future of Free Speech, the Foundation for American Innovation, and the Collective Intelligence Project, to assess how Claude handles contentious political topics. The evaluations found that Claude generally engaged with political questions but sometimes added unsolicited caveats or opposing arguments. Another assessment found that Claude Sonnet 4.5 was more likely to soften arguments supporting right-leaning positions. Anthropic said it has improved how the chatbot presents counterarguments and responds better to direct yes-or-no questions and requests for shorter answers. Anthropic’s Usage Policy prohibits using Claude to spread false voting information, impersonate election officials or suppress voter participation. Researchers found that Claude generally refused direct harmful requests but could miss coordinated campaigns when individual tasks appeared harmless. The company has since introduced additional detection systems targeting fake accounts, misleading news websites, foreign state-linked operations and cyber threats involving election infrastructure. Anthropic said foreign-based operators used Claude to produce US midterm coverage favouring Russian, Iranian and North Korean narratives through a fake news website.
It also identified attempts to generate fabricated news to attract American audiences and earn advertising revenue. The company said it banned the accounts involved and improved its detection systems. Claude.ai will continue directing users to TurboVote, a nonpartisan voting information resource operated by Democracy Works. The chatbot will also begin providing official, state-specific information about polling locations, registration and identification requirements. It also restricts fake social media profiles, mass-generated comments designed to inflate engagement and content falsely presented as independent journalism. Working with the Institute for Strategic Dialogue and the McCain Institute, Anthropic tested its safeguards against election-related misuse in the US, Israel and Brazil. Anthropic plans to continue strengthening election safeguards and publish further analysis of how people use Claude during the election cycle. You use AI every day. Now get your AI Quotient. Take the AIQ test.