The 2024 Election and AI

2024 was the first major election cycle in which generative AI became widely adopted, and also the first election year since Claude launched. Concerned about potential impacts on election outcomes, Anthropic implemented proactive safety measures and used a new analysis tool, Clio, to analyze usage patterns.

Anthropic's safety approach consists of three key components:

• Election-related policies: Established and enforced comprehensive usage policies prohibiting campaigning, election interference, generation of disinformation, and more.
• Rigorous model testing: Conducted red-teaming activities with external experts to test the model's vulnerabilities, verifying disinformation detection and fairness across candidates.
• Providing authoritative information: Directed users seeking voting information to nonpartisan, authoritative sources such as each country's official election authorities.

Over the past year, approximately 100 instances of election-related policy enforcement (such as warnings and account suspensions) occurred worldwide. Unlike social media, chatbots operate through one-on-one interactions, which lowers the risk of information spread, and since Claude currently only outputs text, it is also free from deepfake threats.

Analysis through Clio showed that election-related usage noticeably increased during the week of the US presidential election, but never exceeded 1% of total usage. About two-thirds of users utilized Claude for educational purposes, such as understanding political systems, policies, and analyzing political data.