Adaptive Attack.

An attack on a machine-learning system that is specifically designed to evade or break a known defence, instead of using a generic, defence-agnostic technique. It belongs to the AI & ML Security category of cybersecurity.

An attack on a machine-learning system that is specifically designed to evade or break a known defence, instead of using a generic, defence-agnostic technique.

An adaptive attack is constructed with full knowledge of the target defence and its assumptions, and its loss function or constraints are tailored to bypass that defence. The term was popularised by Carlini and Wagner, whose evaluations repeatedly showed that defences claiming robustness against generic adversarial examples were broken once an attacker designed an objective specifically against them. Adaptive attacks now serve as a standard benchmark: any proposed defence for adversarial examples, watermarking, or detection should be evaluated against adversaries that are aware of the defence and can adapt their methodology. Skipping this step routinely leads to overstated robustness claims that fall to simple, principled attacks.

Defences for Adaptive Attack typically combine technical controls and operational practices, as detailed in the full definition above.

Common alternative names include: Defence-aware attack, White-box adaptive evaluation.