Anthropic released updated election safeguards on October 14, 2026, saying it aims to prevent election-related abuse while supporting political conversations. It is the company's first safety update since its April 2026 announcement on election integrity policies.
Anthropic reported 5.75% of US conversations on Claude.ai involve politics, with 0.51% specifically about elections, measured over a one-week period in September 2026. That compares with 0.01% of conversations asking for voting recommendations.
Claude is built on Anthropic Insights and targets providing accurate information for political discussions. Availability begins with enhanced safeguards for US users, initially for those engaging in political conversations.
"People rarely ask Claude whom or what to vote for," said Anthropic. "More often, they ask questions that help them reach their own conclusions." The company noted that requests violating its election integrity policies were even less common, appearing in about 0.002% of conversations.
The announcement follows a partnership with the Institute for Strategic Dialogue (ISD) to test and strengthen defenses. Anthropic said it is continuing to refine evaluations to better reflect real conversations and cover candidates and voting questions as well as policy issues.
Anthropic did not say how it will handle extreme political scenarios, and raised the open question of whether its models still fall short in some areas. The company said it is addressing these issues through improved testing and responses.
Source: anthropic
