Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
AI Research
arXiv CS.AI·7d ago·5 min read
The AI Risk Observatory: What Can We Learn from AI Disclosures in Annual Reports About Societal Resilience?
TL;DR
A study analyzed 9,821 annual reports from UK companies to assess AI risk disclosures, revealing a significant increase in mentions of AI risk but a lack of substantive information. Only 4.3% of reports contained meaningful AI risk disclosures by 2025.
✦ Why It Matters
Evaluate your company's annual report for AI risk disclosures to ensure transparency and accountability.
Key Takeaways
- Analyze your annual report for AI risk mentions.
- Benchmark against industry standards for AI disclosures.
- Encourage substantive risk disclosures in your organization.
- Monitor sector-specific trends in AI risk reporting.
Full Summary
Research on societal resilience requires actionable data, prompting an investigation into AI risk disclosures in annual reports. Using a two-stage classification pipeline, the study processed 9,821 reports from 1,362 UK companies covering 2020-2025.
Validation against human-annotated passages showed high recall and moderate agreement. Key findings include a rise in AI risk mentions from 2.8% to 41.2% and AI adoption disclosures from 13.8% to 45.2%.
However, disclosures varied significantly by sector, with critical infrastructure sectors lagging. Notably, substantive disclosures about potential harms were almost nonexistent, with only 4.3% of reports containing meaningful AI risk information.
This highlights a gap in transparency regarding AI risks in corporate communications.