Scenario 4: AI Critical Infrastructure Failure Cascade¶

Summary¶

By 2032, AI systems manage much of Australia's critical infrastructure. AEMO uses AI to optimise the National Electricity Market across solar, wind and coal. Sydney Trains runs AI-assisted scheduling and signal control. Telstra and Optus use AI network management for 5G and fibre. Major ports in Melbourne, Sydney and Brisbane rely on AI logistics coordination.

Each system works well independently. Energy costs drop. Train delays decline. Network reliability improves. But the systems are optimised for efficiency under normal conditions—and they're increasingly interdependent.

January 2035. An extreme heatwave hits eastern Australia while geopolitical tensions disrupt coal and gas imports. The electricity grid's AI system, optimised for cost minimisation, makes a series of decisions that seem rational individually but cascade catastrophically:

It shuts down backup capacity that seemed unnecessary. It delays bringing expensive peaker plants online. When demand spikes, it implements rolling blackouts using an algorithm that prioritises industrial users but doesn't account for hospital dependencies or water pumping stations.

The blackouts take down telecommunications towers. Telstra's AI network management system, unable to communicate with affected regions, makes routing decisions that congest the remaining network. Port logistics systems lose real-time data and shut down as a safety measure, halting automated cranes.

Operators try manual overrides, but the AI systems are too complex and interconnected. The override procedures assume access to real-time data that's no longer available. Senior staff who understood the pre-AI systems retired years ago.

The cascade takes 72 hours to contain and two weeks to fully restore. The post-incident review finds that no single system failed catastrophically—but nobody had tested how they would interact under combined stress. Accountability is diffuse: the energy AI was built by a US firm, customised by an Indian contractor and operated by state authorities following federal rules.

What went wrong: C·A·G·R analysis¶

This scenario tests all four pillars simultaneously—making it valuable for comprehensive preparedness exercises.

Questions for actors¶

Use these questions for risk assessments, strategic planning and tabletop exercises.

Why this scenario matters for C·A·G·R¶

This is your "everything fails at once" scenario—the most comprehensive test of defence-in-depth through the C·A·G·R framework. Systems that shouldn't have been deployed with these risks were deployed anyway (Layer 1 fails). AI control methods and regulatory oversight prove inadequate under stress (Layer 2 fails). Resilience measures are the last line of defence (Layer 3 tested). If you can handle this scenario, your preparedness is robust. If you can't, you've identified critical gaps across multiple pillars.