Politicians on AI: A Dataset Tracking 15,000 Official Statements

Summary

The CeSIA and ETH Zurich project “Politicians on AI” tracks 15,000 official statements by political figures about artificial intelligence, allowing users to examine how concerns change over time and differ across countries. It was built and maintained by Lennart Finke of ETH Zurich and Markov Grey of CeSIA, and the page provides an explorer, methodology, definitions, analysis, and downloadable data. The corpus is collected from official government sources including parliamentary records, ministry press pages, and institutional APIs. Documents are split into utterances attributed to one speaker, then filtered for AI mentions using language-specific keyword lists covering the 11 languages represented in the sources. Three LLM-based judges independently assess each surviving utterance; an item is included only when at least two judges agree that it is substantive, belongs to one speaker, and comes from a public official. The tagging scheme draws on existing AI-risk and AI-regulation taxonomies, supplemented by project-defined keywords. Quotes are abridged and checked against their original sources programmatically, but people do not manually review every quote or tag, and speaker profiles are not all human-checked. The researchers caution that the collection is not suitable for comparing jurisdictions because source coverage and inclusion depend on available data and the volume of AI discourse. They consider trends over time within a jurisdiction more interpretable, while noting that delayed publication of some sources, especially US congressional and House hearing records, can make recent months appear artificially sparse. The collection is open source, the data can be downloaded in CSV, TSV, or JSON, and the page identifies the current dataset as version 2.10.0, updated September 22, 2026.