Adversarial Training
A defence technique where a model is trained on both clean and adversarial examples to improve its robustness against adversarial attacks.
In plain language
Making AI tougher by training it on trick examples alongside normal ones. Like a security guard who practices dealing with various break-in scenarios to be better prepared for real ones.
Why this matters
Adversarial training is one of the most practical defences your implementation teams can apply, but it is a trade-off rather than a cure: hardening a model against attacks can reduce its accuracy on ordinary inputs and never makes it fully attack-proof. Governance should ensure these decisions are made deliberately, that the residual risk is documented and that adversarial robustness is re-tested as both the model and the threat landscape evolve.
Relevance
ImplementationAdversarial training is a hands-on defence applied during model development. Engineers weigh its trade-offs and document the residual risk that remains.
Related terms
Putting adversarial training 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.