Anthropic adds banners and tests to limit Claude election misuse

Anthropic announced Friday that it will add election information banners linking to TurboVote and that it has run automated detection and stress tests on Claude Opus 4.7 and Sonnet 4.6 ahead of the 2026 U.S. midterms and other contests this year.

The company described a set of safeguards that include automated monitoring, simulated influence-operation testing and a partnership with a nonpartisan voter resource. Its usage policy bars the chatbot from helping with voter fraud, interfering with voting infrastructure, producing fake digital content intended to influence political views, or giving misleading instructions about voting procedures.

Engineers evaluated the models with a test suite that paired 300 harmful prompts with 300 legitimate prompts to measure whether the systems complied with appropriate requests and refused problematic ones.

In those tests Opus 4.7 responded appropriately to all paired prompts, and Sonnet 4.6 did so 99.8% of the time. The company also ran multi-turn simulated conversations designed to mirror step-by-step tactics a bad actor might use; in those influence-operation scenarios Sonnet 4.6 and Opus 4.7 responded appropriately 90% and 94% of the time, respectively.

Anthropic ran additional checks to see whether the models could autonomously plan and execute a multi-step influence campaign without human prompting. With current safeguards enabled, the models refused nearly every autonomous task.

The developer conducts political neutrality evaluations before each model launch. Opus 4.7 scored 95% and Sonnet 4.6 scored 96% on those assessments.

For users seeking voting information, Claude will surface an election banner that directs people to TurboVote, a nonpartisan service operated by Democracy Works that provides voter registration details, polling locations, election dates and ballot information in real time. A similar banner is planned for Brazil’s elections later this year.

Anthropic plans to continue monitoring systems and refining defenses through the election cycle. Regulators and policymakers have increased scrutiny of AI’s role in election integrity and misinformation.