Breaking the Black Box: Byte-Level Boundary Inference of Real-World Antivirus Systems
AVHunter infers byte-level decision boundaries of 11 antivirus products, exposing compact regions that enable evasion.
Researchers introduce AVHunter, a black-box framework that localizes byte-level decision-critical regions in real-world antivirus products. Probing 11 AV engines produced the Byte-Level AV Boundary Dataset and showed detections often depend on a few compact byte regions. AV-specific models reached 85.07% average boundary-prediction recall and 97.43% agreement with the target AV decisions. Boundary-guided edits evaded detection, induced false positives on benign executables, and remained largely stable across a seven-month study.
- AVHunter probed 11 real-world antivirus products under a black-box model.
- Average boundary prediction recall was 85.07% with 97.43% detection agreement.
- Predicted regions enabled evasion and false positives on benign executables.
- A seven-month study found inferred regions largely stable as AVs evolved.
From the original article
1 min read at arxiv.orgExisting approaches for understanding the detection logic of real-world antivirus (AV) software infer only binary malware/benign decisions from black-box queries, providing limited insight into the fine-grained decision-critical regions that govern AV detection.
By Jieshuai Yang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.