Malware detectors fail on new datasets, impacting enterprise security.

Machine learning models designed to detect malware on Windows systems are usually tested on data similar to their training sets. However, real-world malware on enterprise endpoints often differs, originating from various sources. It is frequently obfuscated to evade detection. A study by Polytechnic of Porto researchers explored this discrepancy. Their findings have significant implications for organizations that depend on static malware detection methods.