Antivirus signatures are the product of expert analysis and encode valuable detection knowledge. But when a system allows black-box queries, attackers can repeatedly probe it to reverse-engineer those signatures, enabling unauthorized reuse or targeted evasion. This exposure undermines the effort invested in building and maintaining detection capabilities.
The paper introduces Signature Knowledge Shielding, a method designed to protect this knowledge. While the abstract does not detail the inner workings, the approach is based on robust embedding matching, presumably to make it harder for query responses to leak useful signature information.
The significance is practical: if effective, the method could help antivirus vendors defend their intellectual property and maintain the integrity of their detection systems without sacrificing usability. The full paper would clarify how the shielding is implemented and evaluated.