Evasion Attack
An adversarial attack performed at inference time where inputs are crafted to evade detection or cause misclassification by a deployed AI model, testing the robustness of the system in production.
In plain language
Crafting inputs to sneak past an AI at the point of use. Like spam emails carefully worded to get past spam filters or malware designed to avoid AI-based security detection.
Why this matters
Evasion attacks are a security governance concern because they reveal vulnerabilities in deployed AI systems that could be exploited maliciously. Your risk assessment should include adversarial robustness testing for any AI system facing untrusted inputs.
Relevance
GovernanceEvasion attack resilience is a security control that organisations should assess for any AI system handling adversarial inputs; governance frameworks should mandate adversarial testing before deployment to production.
Related terms
Putting evasion attack 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.