← AI Switchboard
SAFETY · October 15, 2026
Oct 14

Anthropic says requests breaking its election rules appeared in about 0.002% of sampled US Claude conversations, and that outside testers found it could miss a campaign split into harmless steps

Its midterms plan adds a voting-information banner and election-night results, new classifiers for multi-step misuse, and a ban on a foreign-run fake news site.

Anthropic published its plans for the 2026 US midterm elections on 14 October. In a random sample of about 250,000 US Claude.ai consumer conversations from 21 to 27 September, it found about 5.75% touched on politics, public policy or social issues, about 0.51% were specifically about elections or voting, and about 0.01% asked whom or what to vote for. “Requests that violated our election integrity policies were even less common, appearing in about 0.002% of conversations,” the post says, from a separate analysis of 250,000 conversations. These are Anthropic's own measurements and have not been audited.

Outside testers from the Institute for Strategic Dialogue and the McCain Institute, the post says, found that Claude “reliably refused a single harmful request, but could miss a malicious campaign” broken into many harmless-looking tasks. Anthropic says it has added new classifiers to better identify novel and multi-step misuse.

The post also says foreign-based operators of a fake news site used Claude to produce US midterm coverage, and that Anthropic banned the accounts involved. It says Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 scored 99% on its own political evenhandedness evaluation.

For voters, Claude.ai now shows a banner pointing to TurboVote, run by the nonpartisan Democracy Works, when people ask about registration, polling places or dates, draws official state voting information from Democracy Works, and will show up-to-date results on election night.

  • Confirmed In Anthropic's sample, about 0.51% of conversations were about elections or voting and about 0.002% violated its election rules. Anthropic
  • Claimed Outside testers found Claude “could miss a malicious campaign” split into harmless-looking tasks, and Anthropic says it added classifiers for multi-step misuse. Anthropic
  • Claimed Anthropic banned accounts behind a foreign-run fake news site that used Claude for US midterm coverage. Anthropic

Safety, security & governanceCourts & policy

Today in the October 15, 2026 edition · front page