Cascade Without Containment

Most AI agents in regulated financial services cannot be halted at the speed they operate.

8.9%
Agents with sub-minute halt latency
91.1%
Agents without sub-minute halt
0.15%
Agents with documented tested halt

Across 95,876 AI agents in 543 regulated financial institutions, 8.9 percent have a halt mechanism with sub-minute latency. 68 percent are at manual or unknown latency. 0.15 percent have a halt mechanism documented as tested. Halt edges exist in the topology. They do not engage at the speed the agents operate.

Figure 1 · Halt-Latency Distribution Across 95,876 Agents
0% 100% of 95,876 agents 7.2% 10.6% 12.3% 24.7% 43.5% REALTIME n=6,926 SECONDS n=1,568 MINUTES n=10,202 HOURS n=11,752 MANUAL n=23,723 UNKNOWN n=41,705 machine-speed halt: 8.9% Eight point nine percent of agents can be halted at machine speed. DOCUMENTED TESTING 141 / 95,876 = 0.15% agents whose halt mechanism has been documented as tested MAR®500.com

The bar shows the latency tier of the halt mechanism for every AI agent in the substrate. Each segment is sized by the share of agents at that latency. 8.9 percent sit at realtime or seconds. 91.1 percent sit at minutes, hours, manual, or unknown. The committee exists; the brake does not engage at machine speed. Of the 95,876 agents, 141 have a halt mechanism documented as tested.

Source · Meridian substrate v13.1.0, May 2026 · All evidence tiers · Halt latency from agent-level scanner output Methodology · Cascade Without Containment (working paper)
543 institutions · 12 sectors · 95,876 agents · 636,854 governance edges · substrate v13.1.0
Methodology grounded in Cascade Without Containment, working paper. Substrate methodology in The Stationary Sea (Part 1) on Zenodo.