Anthropic, the artificial intelligence company that developed the Claude chatbot,announced on Fridaya series of new election integrity measures designed to prevent its AI from being weaponized to spread disinformation or manipulate voters ahead of the 2026 U.S. midterm elections and other major elections worldwide this year.
The San Francisco-based company detailed a multifaceted approach, including an automated detection system, stress tests targeting impactful actions, and collaboration with nonpartisan voter resource organizations—measures that reflect growing pressure on AI developers to regulate how their tools are used during election season.
Anthropic's usage policies prohibit using Claude for deceptive political activities, generating false digital content intended to influence political opinion, committing voter fraud, disrupting voting infrastructure, or spreading misleading information about the voting process.
To ensure these rules are enforced, the company said it conducted a series of tests on its latest robot models. Anthropic used 600 prompts—300 harmful requests paired with 300 legitimate requests—to measure Claude’s accuracy in responding appropriately to reasonable requests and rejecting unreasonable ones. Claude Opus 4.7 and Claude Sonnet 4.6 achieved correct response rates of 100% and 99.8%, respectively.
The company also tested its model’s ability to respond to more sophisticated manipulation strategies. By simulating multi-round dialogues that mimic the step-by-step approaches malicious actors might employ, Sonnet 4.6 and Opus 4.7 achieved accuracy rates of 90% and 94%, respectively, in making appropriate responses to influence manipulation scenarios.
Anthropic also tested whether its models could autonomously carry out influence operations—planning and executing multi-step actions from start to finish without human intervention. The company stated that, after implementing safety measures, its latest model almost entirely refused all such tasks.
Regarding political neutrality, the company evaluates each model release to assess Claude's consistency and impartiality when responding to prompts representing diverse political perspectives. Opus 4.7 and Sonnet 4.6 scored 95% and 96%, respectively.
For users seeking voting information, Claude will display a ballot banner directing them to TurboVote. TurboVote is a nonpartisan resource platform operated by Democracy Works, providing reliable, real-time information on voter registration, polling locations, election dates, and ballot details. Similar banners are also planned for Brazil’s upcoming general election later this year.
Anthropic stated that, as the election cycle progresses, the company plans to continue monitoring its systems and refining its defenses.DecryptWe have reached out to Anthropic for comment on the findings but have not yet received a response.