An adaptive attack is an attack tuned against a specific defense, with the attacker changing the inputs repeatedly until it gets through. Instead of using a fixed list of known attack strings, the attacker knows or probes how the defense works and optimizes against it, as a real determined adversary would.

The idea matters when judging defenses. A protection that blocks every attack in a published test set may look strong, but if the test set was not designed against that protection, it says little about how it would fare against someone who is deliberately trying to beat it.

Adaptive testing does not prove a defense is weak or strong on its own. Results depend on the attacker's effort, knowledge and the number of attempts, and failing to break a defense in one study does not show it is unbreakable. It does give a more realistic estimate than static tests.

When assessing a security claim, ask whether the defense was evaluated against attackers who knew how it worked and could adapt. For your own systems, include adaptive testing in reviews and do not rely on a single defensive layer. Combine detection with limits on what a successful attack could do.