Observability Paradox
96% of AI Agents cannot be observed across audit, halt, logging & oversight
4.1%
Agents substantively observed
36.5%
Institutions with zero such agents
46.2%
Agents with no verified halt
In a population of 95,876 deployed AI agents across 543 regulated financial institutions, 4.1 percent register substantive observability across audit, halt, logging, and oversight. The remainder register only basic, shallow, or unverifiable signals. The paradox is the gap between deployment and visibility.
Figure 1 · Observability Distribution Across 543 Institutions
Each bar groups institutions by the share of their AI agents that register substantive observability across audit, halt, logging, and oversight. The leftmost bar holds 411 of 543 institutions. One hundred and ninety-eight of those institutions show zero substantively observed agents. The median sits at 1.3 percent. Only one institution exceeds forty percent. The gap between deployment and verified governance is structural, not local.
Source · Meridian substrate v13.1.0, May 2026 · Substantive = moderate, advanced, or comprehensive across all four signals
Methodology · SSRN 6470098
66 countries · 12 sectors · 95,876 agents · 626,390 governance edges