The inadequacy of entropy-based ransomware detection

Many state-of-the-art anti-ransomware implementations monitoring file system activities choose to monitor file entropy-based changes to determine whether the changes may have been committed by ransomware, or to distinguish between compression and encryption operations. However, such detections can be victims of spoofing attacks, when attackers manipulate the entropy values in the expected range during the attacks. This paper explored the limitations of entropy-based ransomware detection on several different file types. We demonstrated how to use Base64-Encoding and Distributed Non-Selective Partial Encryption to manipulate entropy values and to bypass current entropy-based detection mechanisms. By exploiting this vulnerability, attackers can avoid entropy-based detection or degrade detection performance. We recommended that the practice of relying on file entropy change thresholds to detect ransomware encryption should be deprecated.