Researchers at the University of Texas at Dallas say they have created a piece of malware that can pull together normally benign code from other programs to avoid detection.
The researchers – Vishwath Mohan and Kevin W. Hamlen – are calling the malware ‘Frankenstein’, after the fictional character of the same name. According to a paper by the duo, Frankenstein changes the game for authors of metamorphic malware who are accustomed to rewriting code so that each variant is different from the last. Instead, Frankenstein uses instructions from non-malicious programs that are considered safe by security programs, which the researchers contend makes it more difficult for “feature-based malware detectors to reliably use those byte sequences as a signature to detect malware.”
“The instruction sequence harvesting process leverages recent advances in gadget discovery for return-oriented programming,” according to a paper the researchers released describing their findings. “Preliminary tests show that mining just a few local programs is sufficient to provide enough gadgets to implement arbitrary functionality.”
The research was funded in part by the U.S. Air Force Office of Science Research. The two detailed their findings at the 6th USENIX Workshop on Offensive Technologies (WOOT) in Bellevue, WA, earlier this month.
“Rather than recompiling the code purely randomly during propagation, which leads to diverse but potentially distinguishable binary features, our system searches non-malicious programs on the local system for byte sequences that function as the building blocks for semantically equivalent but syntactically new copies,” the researchers noted in the paper. “Our experiments showed that mining a few files is both sufficient to obtain high mutant diversity, and fast enough to be a practical mutation strategy.”
“By creating new copies entirely from byte sequences obtained from benign files,” the paper continues, “we argue that it becomes significantly more difficult for defenders to infer adequate signatures that reliably distinguish malware from non-malware on victim systems. In particular, signatures that include feature-whitelisting are less effective against our framework than against more conventional forms of obfuscation.”
The paper can be downloaded here.