Fail-Safe
A design principle ensuring that an AI system defaults to a safe state or ceases operation when a failure, error or anomalous condition is detected.
In plain language
Designing AI to default to a safe state when things go wrong. If a self-driving car's AI cannot figure out what to do, it should pull over and stop, not speed up.
Why this matters
Fail-safe design is a governance principle that organisations should apply to high-risk AI systems. Requiring systems to default to a safe state rather than attempting uncertain actions reduces the likelihood of cascading failures and supports compliance with operational safety standards.
Relevance
GovernanceFail-safe design is a control mechanism that governance frameworks should mandate for systems making consequential decisions affecting people or assets.
Related terms
Putting fail-safe into practice in your organisation?
Ready to transform your AI strategy?
Partner with Australia's AI strategy and governance specialists. From adoption roadmaps to ISO 42001 audit readiness.