The Latency Gap

Even at the highest risk tier, halt latency does not reach machine speed.

13.0%
High-risk agents with sub-minute halt
6.4%
Other agents with sub-minute halt
57.7%
High-risk agents at manual or unknown

35,398 AI agents in the substrate are flagged as EU AI Act high-risk by attribute match. 13.05 percent of these have a halt mechanism with sub-minute latency. 57.72 percent are at manual or unknown latency. Among the remaining 60,478 agents in the substrate, sub-minute coverage drops to 6.41 percent. The risk tier moves the needle, but not across the threshold.

Figure 1 · Halt-Latency Stratified by EU AI Act High-Risk Status
HIGH-RISK n=35,398 11.1% 12.4% 16.9% 25.0% 32.7% sub-minute: 13.0% OTHER n=60,478 5.0% 9.6% 9.6% 24.6% 49.8% sub-minute: 6.4% REALTIME SECONDS MINUTES HOURS MANUAL UNKNOWN Even at the highest risk tier, more than half of agents lack sub-hour halt latency. MAR®500.com

The two bars show the latency distribution of halt mechanisms for two cohorts: agents flagged as EU AI Act high-risk (top, n=35,398) and the remaining agents (bottom, n=60,478). Each segment is sized by the share of agents at that latency. The dashed gold line marks the sub-minute threshold. High-risk agents do receive more attention. Sub-minute halt coverage is twice as high. But more than half remain at manual or unknown latency. The risk classification moves resources, not the threshold.

Source · Meridian substrate v13.1.0, May 2026 · EU AI Act high-risk = yes or likely · All evidence tiers 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.