Executive brief
AgentVerus Scanner is a security analysis tool designed to detect malicious code in AI agent skills. The scanner fails to inspect compiled Python bytecode files stored in __pycache__ directories, allowing attackers to distribute malicious compiled code alongside legitimate source files. This enables arbitrary code execution while the scanner incorrectly reports the code as safe and trusted.
Technical details
The vulnerability is a file analysis bypass in AgentVerus Scanner where the analyzer fails to scan compiled Python bytecode (.pyc) files stored in __pycache__ companion directories. An attacker can craft malicious Python bytecode that executes arbitrary code on import while the scanner's static and semantic analysis components report a CERTIFIED verdict. The attack requires no authentication and exploits the scanner's assumption that scanning source files is sufficient, failing to recognize that Python will execute the precompiled bytecode in preference to the source. No patch status was disclosed in the advisory.
Affected products
- AgentVerus Scanner 0.8.1 and earlier
Timeline
- 2026-09-02: disclosed