What it means
Where multiple AI systems interact, one agent’s output becoming another’s input; an error, attack, or misalignment can propagate and amplify across interconnected systems before anyone sees it. This is categorically different from a single AI failing: the scale of consequence is determined by the interconnectedness of the architecture, not the severity of the initial failure.
Why it matters
In financial markets, supply chains, infrastructure, or enterprise operations, cascading AI failures can produce systemic consequences that no individual system failure would have caused. The speed of propagation can exceed the speed of human response.
Board governance implications
The board must understand the points of interconnection between AI systems in the organisation’s architecture and confirm that human override mechanisms exist at each. Incident response planning must include cascade failure scenarios, not just single-system failures.
Governance failure timeline
Pre-deployment
Failure to map AI system interconnections, establish human override mechanisms at each point of integration, and include cascade failure scenarios in incident response planning before interconnected systems are approved for live use.
Deployment
Interconnected AI systems are operating without override mechanisms or propagation monitoring.
A single failure point is capable of amplifying to systemic consequences before any human response is possible.
The speed of propagation is determined by the architecture, not the severity of the initiating failure.
Post-deployment
Systemic failures have produced financial, operational, reputational, and regulatory consequences at a scale no individual AI failure would have caused.
The Board must account not only for the initiating failure but for the absence of the interconnection mapping, override mechanisms, and cascade scenario planning that would have contained it.