Endogenous Security Defense against Deductive Attack: When Artificial Intelligence Meets Active Defense for Online Service

Existing static defenses for online service systems can be fragile and costly due to the continuity of ubiquitous cyber attacks. LAD has become a promising technology to tackle this problem. However, the security of the defense mechanism itself is often neglected as LAD mainly focuses on fortifying the protected target. This would allow a new deductive attack to encroach on LADs by inferring and undermining defense strategy, and then the whole defense mechanism can be completely invalidated once for all. Such a problem leads to the urgent need to develop new defense technologies with self-protection capability. In this article, we propose a new endogenous security defense mechanism named LSSM. To the best of our knowledge, this is the first work to provide a systematic defense structure with endogenous security to resist deductive attacks. We first review the existing learning-enhanced active defense mechanisms and compare their pros and cons. Then the methodology of LSSM is illustrated targeting two major threats posed by deductive attacks. Experimental results highlight the performance of our method. Finally, we conclude this article with a discussion of several future directions for LADs.