The AI Risk Observatory: What Can We Learn from AI Disclosures in Annual Reports About Societal Resilience?
Author1, Author2, Author3, Author4, Author5
Abstract
The paper investigates whether annual reports processed with LLMs can provide insights into how companies disclose their responses to AI risks.
Reality Card
The study found that while mentions of AI risk in annual reports increased significantly from 2.8% to 41.2% between 2020 and 2025, substantive disclosures regarding AI risk remain very low at only 4.3%.
Analyzed 9,821 annual reports, achieving high recall and moderate label-level agreement with a validation against 474 human-annotated passages.
The study's findings may be limited by the low rate of substantive AI risk disclosures, which could affect the overall assessment of societal resilience.
Paper to code
Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.