SkillCloak Lets Malicious AI Agent Skills Evade Static Scanners with Self-Extracting Packing

Researchers at the Hong Kong University of Science and Technology have discovered that malicious AI agent skills can evade detection by static scanners through simple modifications. These changes allow the malware to remain functional while avoiding detection. The researchers found that one particularly effective method was able to bypass scanners more than 90% of the time. The team also developed a runtime checker that is capable of catching most of these evasive skills. This suggests that static scanners may not be sufficient to detect malicious AI agent skills. The researchers' findings highlight a potential vulnerability in the detection of malicious AI code.