MEG Case Studies

Welcome to the MEG (Minimal Ethical Governance) Case Studies Index. This compendium serves as an operational and technical-legal testbed for the MEG framework (incorporating both the technical specifications of MEG 1 and the legal liability models of MEG 2).

To bridge the "Accountability Gap" in modern and Agentic AI systems, we subject real-world failures, catastrophic model drifts, and systemic historical incidents to rigorous normative evaluations under the MEG protocol.

The index below contains case studies:

• Cases 001 – 010: Modern and complex Agentic AI failures, platform security exploits, and commercial liability developments from 2024 to 2026 (including Replit, PocketOS/Cursor, and the Munich Re/HSB actuarial integrations).
• Cases 011 – 110: A comprehensive compendium of historical AI and algorithmic incidents spanning facial recognition abuses, automated benefit fraud rejections, conversational harms, and predictive policing failures. Each historical case includes a dedicated analysis of "How MEG 1 would have acted" to prevent or mitigate the harm.

New cases (N-series) are documented after the framework publication and are maintained as a living annex. Unlike cases 001–110, they are not yet included in the archived Zenodo release and carry no persistent DOI; each is labelled Grounded, Probed or Analogical to indicate evidentiary status.

For detailed technical schemas, persistent DOIs on Zenodo, and full framework specifications, please consult our primary documentation library:

Primary Documentation & Resources:
• Specification Library: meg-initiative.org/library/
• MEG1: A Technical Standard for Verifiable AI (Zenodo): doi.org/10.5281/zenodo.21280679
• MEG2: Legal Governance for AI (Zenodo): doi.org/10.5281/zenodo.21280675

Select any case below to see exactly how MEG 1 + MEG 2 would have prevented, mitigated, or allocated liability for the incident.