Think of the last time you flew commercially. As your plane taxied to the runway, you probably didn’t worry about engine failure, navigation errors, or pilot fatigue – because aviation has built multiple independent safety barriers that work together to prevent incidents. This approach, known as the Swiss cheese model, visualizes each safety layer as a slice of Swiss cheese, with holes representing potential failures or weaknesses.
This same model that has made commercial aviation extraordinarily safe is exactly what your organisation needs for AI deployment. The layers would typically include:
- Human factors – Supervision and training on risks, scope and usage, operations management, and standardised procedures that reduce the likelihood of human error.
- Technical systems – Multiple redundant infrastructure systems, regular backups, historical event logs, data management systems and automated alerting.
- Organisational controls – Deployment management systems, clear policies and guidelines, regular audits, incident reporting culture, and continuous improvement processes that identify and address systemic risks.
- Regulatory oversight – Standards, governance frameworks, regular human review, and automated enforcement mechanisms.
The key insight is that no single layer is perfect – each has “holes” where failures can occur. However, when properly implemented, the holes rarely align across all layers simultaneously. An incident typically requires multiple concurrent failures – the holes must line up for a hazard to pass through all defences.
Your success in deploying AI comes from continuously monitoring each layer, learning from training and near-misses, and strengthening weak points before they contribute to a public incident.
This systematic approach has made commercial aviation extraordinarily safe, and that same methodology can be used to ensure your AI is just as safe.