Most companies mention AI risk without substantively addressing it
AI risk appeared in 41.2% of UK listed companies’ annual reports in 2025, but only 4.3% contained disclosure classified as substantive, giving resilience researchers a large signal with little evidence of depth. The study uses an LLM pipeline to classify 9,821 reports from 1,362 companies, validated against 474 human-annotated passages. Harm disclosures were especially rare: just seven reports across the corpus.