Anthropic gears up Claude for the US midterm elections

Anthropic is stepping up efforts to make its Claude AI chatbot ready for the upcoming US midterm elections, with a stronger focus on election integrity and preventing the misuse of artificial intelligence. The company says the new measures are designed to limit political manipulation, election-related misinformation and coordinated attempts to influence voters.

As part of the changes, Claude will provide users with official voting information, including state-specific details on registration, polling locations and identification requirements. The chatbot will continue directing users to TurboVote, a nonpartisan voting information service operated by Democracy Works. Anthropic also plans to provide updated election results and vote counts during election night and in the days following the polls.

The company studied around 250,000 Claude conversations over a one-week period to understand how Americans use the chatbot for political topics. Politics, public policy and social issues accounted for 5.75 per cent of conversations on Claude.ai in the US, while queries specifically related to elections and voting made up 0.51 per cent.

Requests asking Claude to recommend whom or what to vote for represented about 0.01 per cent of conversations. Meanwhile, queries that violated Anthropic’s election integrity policies accounted for roughly 0.002 per cent.

Focus on political neutrality

Anthropic is also strengthening its efforts to make Claude more even-handed when handling political discussions. The company said Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 scored 99 per cent in its political even-handedness evaluation.

The company worked with groups including Vanderbilt University's Future of Free Speech, the Foundation for American Innovation and the Collective Intelligence Project to examine how Claude responds to controversial political subjects.

The assessments found that Claude generally engaged with political questions but could sometimes introduce unsolicited caveats or opposing viewpoints. One evaluation also found that Claude Sonnet 4.5 was more likely to soften arguments supporting right-leaning positions. Anthropic said it has since improved the model's handling of counterarguments and its ability to answer direct yes-or-no questions and shorter requests.

Tighter safeguards against election abuse

Anthropic’s Usage Policy prohibits using Claude to distribute false voting information, impersonate election officials or discourage people from voting. It also bans activities involving fake social media accounts, mass-produced comments intended to artificially boost engagement and material falsely presented as independent journalism.

Tests conducted with the Institute for Strategic Dialogue and the McCain Institute examined potential election-related misuse in the US, Israel and Brazil. While Claude generally rejected clearly harmful requests, researchers found that coordinated campaigns could sometimes go undetected when individual tasks appeared harmless on their own.

Anthropic has responded by adding detection systems aimed at identifying fake accounts, misleading news websites, foreign state-linked operations and cyber threats targeting election infrastructure.

The company also said foreign-based operators had used Claude to create US midterm election coverage favouring Russian, Iranian and North Korean narratives through a fake news website. It identified other attempts to generate fabricated news aimed at American audiences and advertising revenue.

Anthropic said it banned the accounts involved and strengthened its detection systems. The company plans to continue improving Claude’s election safeguards and publish more findings throughout the election cycle.

ABOUT THE AUTHOR

Kahekashan

[email protected]

Kahekashan is a passionate technophile with a keen eye for cutting-edge gadgets, emerging technologies, and everything in the digital realm. Raised in a Defence family with strong values and a background in literature, she has consistently pursued excellence in every endeavour. Her last full-time assignment involved content writing with the Indian School of Business.

Read More