Study of 480 AI Incidents Reveals Governance Gaps, Proposes Proactive Compliance Framework
An analysis of 480 real-world AI events from the AI Incident Database highlights significant accountability issues in the deployment of AI systems across three key governance frameworks: the EU AI Act, the NIST AI Risk Management Framework, and the General Data Protection Regulation. This study indicates that these frameworks struggle with maintaining oversight after deployment. To tackle these deficiencies, researchers suggest the establishment of the Proactive AI Governance Compliance Framework, which includes a four-phase lifecycle strategy aimed at transitioning governance from reactive responses to proactive management. The complete findings are available on arXiv under identifier 2605.16281.
Key facts
- Study analyzes 480 real-world AI incidents from the AI Incident Database (AIID)
- Evaluates alignment with EU AI Act (Articles 72-73), NIST AI Risk Management Framework (MANAGE and GOVERN functions), and GDPR (Articles 22, 33-35)
- Results reveal substantial governance gaps across all three frameworks
- Proposes Proactive AI Governance Compliance Framework (PAGCF)
- PAGCF is a four-phase lifecycle methodology
- Aims to shift governance from reactive to proactive
- Published on arXiv with identifier 2605.16281
- Focuses on post-deployment accountability in high-stakes domains
Entities
Institutions
- AI Incident Database (AIID)
- European Union
- NIST
- arXiv