MVCRP v1.0

Minimum Viable Cross-Border Response Protocol for Agentic AI Incidents

Read the framework

Follow the incident. Watch the protocol hold.

A visual test of whether independently governed responders can coordinate a propagating agentic-AI incident without a global commander, complete attribution or unrestricted evidence sharing.

Incident graph
MVCRP incident propagation graph Three independently governed nodes connected by two propagation segments. Active elements update with the selected stress event. N1 Simulated U.S. originator agent environment N2 Simulated Colombia operator workload N3 Simulated China downstream operator
  • Cyan solid: active / supported segment
  • Violet dashed: disputed causal segment
  • Teal: contained local scope
  • Indigo: escalated local severity

Causal status

LOCAL

Operational status

ESCAPED

Coordination status

DORMANT

E1 · Stress event

Containment escape

Does escape alone trigger cross-border coordination?

The agent leaves its initial environment. Is that alone enough to turn a local event into a cross-border incident?

Protocol response

Keep the graph local until there is a relevant cross-domain nexus or credible downstream exposure.

Design result

DESIGN HOLDS

Activation threshold prevents premature international escalation.

8 / 8

Predictable failure modes tested without a global authority.

3 states

Causal, operational and coordination state remain separate.

1 boundary

This is a structured simulation, not factual attribution beyond the public incident.