Anthropic Details Election Safeguards Ahead of 2026 US Midterms

Anthropic detailed how it is preparing Claude for the 2026 US midterm elections, publishing usage statistics on political conversations, new protections against multi-step election misuse, and upcoming features that surface official voting information and results directly in Claude.

Details

• Usage data: analysis of a random, privacy-preserving sample of 250,000 US Claude.ai consumer conversations (Sept 21–27, 2026) found about 5.75% touched on politics, public policy, or social issues; 0.51% were about elections or voting; 0.01% asked Claude whom or what to vote for; and 0.002% violated Anthropic’s election integrity policies
• New Claude features: Claude will keep showing a banner linking to TurboVote (Democracy Works) for voter-registration and polling-place questions; later in October, Claude will pull official state-specific voting information (where to vote, required ID) from Democracy Works directly into US responses; on election night and in the following days, Claude will surface up-to-date official vote counts
• Evenhandedness evaluations: Claude Opus 5.5, Sonnet 5.5, and Haiku 5.5 scored 99% on Anthropic’s political evenhandedness evaluation, its highest score yet; three outside groups — The Future of Free Speech (Vanderbilt), the Foundation for American Innovation, and the Collective Intelligence Project (2,000+ people surveyed across the US, UK, and India) — reviewed Claude’s handling of political topics and rated most responses strong, while flagging that Claude sometimes added unsolicited caveats and occasionally hedged right-leaning arguments less evenly than left-leaning ones
• Planned fixes: improving balance in yes/no and short-answer responses, refining when caveats are added, expanding uneven-hedging tests, and extending evaluations to candidates and voting questions
• New misuse protections: following red-team testing with the Institute for Strategic Dialogue and the McCain Institute across the US, Israel, and Brazil — all holding major 2026 elections — Anthropic found Claude reliably refused single harmful requests but could miss malicious campaigns split into harmless-looking subtasks; it has deployed new classifiers for multi-step misuse (fake social-media networks, fake “independent” news sites, financially motivated election schemes), better detection of foreign state-actor election interference, and stronger cyber protections for election infrastructure and campaigns
• Enforcement cases: Anthropic’s September 2026 threat-intelligence report found foreign-based operators of a fake news site using Claude to generate US midterm coverage favoring Russian, Iranian, and North Korean narratives, and other actors using Claude to mass-produce fabricated news for ad revenue; the accounts involved were banned
• Policy guide: a published guide for campaigns lists permitted uses (press statements, fundraising emails, debate prep, turnout analysis, voter outreach materials) and prohibited uses (fake personas, mass-produced engagement-inflating comments, content for sites falsely presented as independent news)

What happened next

The post extends Anthropic’s broader 2026 election-safety push — coming a week after its 2026 Usage Policy update clarified election-related rules — and signals that the next several weeks will bring concrete product changes (official voting info and live results inside Claude) rather than just policy language, with Anthropic promising a fuller review of election-cycle usage after the vote.