# Anthropic Says Claude Is Safe for Elections. The Hard Numbers Are Buried in the Methodology. - Date: 2026-04-26 - Category: Artificial Intelligence The AI refused every harmful prompt in a one-off test. The test designed to look like a real attack succeeded 6-10% of the time. ---