Anthropic published an update (Oct 14, 2026) on its election safeguards ahead of the 2026 US midterms. Using Anthropic Insights, it analysed about 250,000 US Claude.ai conversations from Sept 21-27, 2026. Per Anthropic: conversations touching politics, public policy or social issues were about 5.75% of US conversations; about 0.51% were specifically about elections or voting; requests asking whom or what to vote for were about 0.01%; and requests that violated its election-integrity policies were about 0.002%. The company says it will keep showing a TurboVote banner from Democracy Works, and that later this month Claude will pull official voting information from Democracy Works directly into answers -- state-specific guidance on where to vote and what ID to bring -- and will show up-to-date election-night results. It also describes a political-evenhandedness evaluation and work with outside groups such as the Institute for Strategic Dialogue and the McCain Institute.
Our takeTwo numbers stand out: politics is a genuine slice of US usage (5.75%), yet explicit 'who should I vote for' asks are vanishingly rare (0.01%), and policy-violating ones rarer still (0.002%). That undercuts the alarmist premise that chatbots mainly whisper voting advice. The open question is whether pulling official results and voting logistics into the answer box makes Claude a de facto civic utility -- and who is accountable when a state's data lags.