AI Risk Observatory: Analyzing AI Disclosures in Annual Reports for Societal Resilience
Summary:This research investigates whether processing annual reports with Large Language Models (LLMs) can provide actionable insights into how companies disclose their response to AI and its implications for societal resilience. Analyzing 9,821 annual reports from 1,362 UK-listed companies (2020-2026), the study reveals that the share of reports mentioning AI risk increased from 2.8% in 2020 to 41.2% in 2025, while AI adoption disclosure rose from 13.8% to 45.2%. However, most AI risk disclosures lack
Background and Motivation
As AI technologies advance, societal reliance on AI increases, making its impact on societal resilience more pronounced. However, the extent to which companies disclose their AI strategies and potential risks remains unclear. This study aims to assess whether AI-related disclosures in annual reports can provide insights into companies' approaches to AI and their implications for societal resilience.
Methodology
The study employs a reproducible two-stage classification pipeline to analyze 9,821 annual reports from 1,362 UK-listed companies (2020-2026).
- Data Processing and Classification: Preprocess reports using LLMs and apply a two-stage classification model to identify AI risk and AI adoption disclosures.
- Validation and Evaluation: Validate the method against 474 human-annotated passages, achieving high recall and moderate label-level agreement.
Key Findings
-
AI Risk Disclosure Trends:
- The share of reports mentioning AI risk increased from 2.8% in 2020 to 41.2% in 2025.
- AI adoption disclosure rose from 13.8% to 45.2%.
- A small set of major providers, led by Microsoft, dominate the vendor mentions.
-
Industry and Market Segment Differences:
- AIM market reports disclose AI risk at much lower rates than Main Market reports.
- Sectors like Energy and Data Infrastructure lag behind in AI risk disclosure.
-
Scarcity of Substantive Disclosures:
- While 41.2% of reports mention AI as a risk, only 4.3% contain disclosures classified as substantive.
- Harm disclosures are near-absent, with only seven reports mentioning them.
Industry Implications and Recommendations
-
For Businesses:
- Increase the substantive content of AI risk disclosures to enhance transparency and credibility.
- Strengthen AI risk assessment and management, especially in critical infrastructure sectors.
-
For Regulators:
- Develop stricter AI disclosure guidelines to ensure companies fully disclose AI-related risks and mitigation strategies.
- Promote the standardization of AI risk assessment criteria to facilitate cross-industry comparability.
-
For Researchers:
- Further investigate the factors and drivers of AI risk disclosure.
- Develop more advanced AI text analysis tools to improve the accuracy and efficiency of disclosure analysis.
— END —Source: ArXiv AI (cs.AI) (2026-10-06)
Tags: #AI Risk #Annual Reports #Societal Resilience #LLMs & Foundation Models #Data Disclosure
Editorial & Fact-Checking Note: This article is compiled from primary research, official release documentation, and source papers by the ZICQ Newsroom pipeline with automated entity verification and human editorial review. If you notice any technical inaccuracy, please submit a correction via our corrections policy or email our editorial desk directly.
Community Comments