Anthropic prepares Claude for the 2026 US midterms

Your AI and this story

Key facts

• Usage data (random sample of 250,000 US Claude.ai consumer conversations, Sept 21–27, 2026, via Anthropic Insights): politics, public policy or social issues about 5.75% of conversations; elections or voting about 0.51%; asking whom or what to vote for about 0.01%; election-integrity policy violations about 0.002% (separate 250,000-conversation sample)
• New features: ‘Later this month’ Claude.ai will draw official voting information from Democracy Works into answers for US users (state-specific polling places, ID rules), adding to the TurboVote banner; Claude will show up-to-date official results on election night and the following days
• Bias: Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 score 99% on Anthropic’s political evenhandedness evaluation, ‘our most evenhanded to date’
• Outside reviews: The Future of Free Speech (Vanderbilt), the Foundation for American Innovation and the Collective Intelligence Project (2,000+ people in the US, UK and India). Findings: Claude rarely refused but sometimes added unsolicited caveats and counterarguments (e.g. an opposing view in an advocacy email); FAI found Claude Sonnet 4.5 ‘more likely to soften arguments for right-leaning positions’
• Red-teaming with the Institute for Strategic Dialogue and the McCain Institute in the US, Israel and Brazil: Claude ‘reliably refused a single harmful request, but could miss a malicious campaign when it was spread across many individual requests that looked harmless on their own’; new classifiers target multi-step misuse, fake accounts and news sites, and financially motivated schemes
• Enforcement: ‘foreign-based operators of a fake news site used Claude to produce US midterm coverage favoring Russian, Iranian, and North Korean narratives’; other actors made fabricated news for ad revenue; the accounts were banned
• Anthropic also published a guide for political campaigns on allowed uses (press statements, debate prep, turnout analysis) and prohibited ones (fake profiles, mass comments, sites falsely claiming independence)