Vendor Governance Dependency

Foundation model vendors generate halt evidence that hyperscalers do not.

1.16×
FMV halt-robust advantage over hyperscalers
76.8%
FMV agents with known halt mechanism
73.8%
Hyperscaler agents with known halt mechanism

The Coase Inversion described in Paper 1 surfaced a structural property of agent governance: a non-trivial share of the halt evidence visible in public sources is generated by the vendor wrapper rather than by the institution itself. We test this property at the agent level by comparing halt evidence between agents served through foundation model vendors and agents served through hyperscale infrastructure providers. Under the v13.1.0 substrate, FMV-served agents exhibit halt knowledge in 76.8 percent of cases against 73.8 for hyperscaler-served, and robust halt categories in 7.29 percent against 6.29. The direction confirms the Paper 1 finding; the magnitude is materially smaller than the v11.1 baseline indicated. The implication for cross-institutional propagation persists: when the vendor changes a halt-relevant configuration, the governance posture changes simultaneously across every dependent institution.

Figure 1 · Halt Evidence by Vendor Type
0% 25% 50% 75% 100% PERCENTAGE OF AGENTS 76.8% 73.8% Halt mechanism known FMV 1.04× Hyperscaler 7.29% 6.29% Halt mechanism robust (moderate or advanced) FMV 1.16× Hyperscaler FMV agents (n=2,840) Hyperscaler (n=7,395) Halt evidence still favours foundation models, narrowly. MAR®500.com

Each pair of bars compares the share of agents with the named halt evidence category between those served through foundation model vendors and those served through hyperscale infrastructure providers. FMV-served agents exhibit modestly higher halt evidence on both metrics, approximately 1.16 times the rate observed on hyperscaler-served agents for robust halt categories. The Paper 1 mechanism, that vendor platform wrappers contribute observable governance evidence beyond what the institution itself generates, holds in direction. The magnitude is narrower than the v11.1 baseline suggested. The cross-institutional implication persists: a vendor configuration change propagates a governance change to every dependent institution at the same instant.

Source · Meridian substrate v13.1.0, May 2026 · Vendor classification from agent-level vendor field; halt categories from agent-level halt field Methodology · SSRN 6535599
543 institutions · 12 sectors · 95,876 agents · 636,854 governance edges · substrate v13.1.0
Methodology grounded in Systemic Governance Risk in AI Agent Networks, SSRN Working Paper 6535599. Substrate methodology in The Stationary Sea (Part 1: Substrate Construction) on Zenodo.