Security Research

GuardFall: Old Bash Parsing Lets Malicious Commands Slip Past AI Agent Guards

Researchers say a mismatch between text-based checks and Bash’s runtime rewriting can let agents run destructive shell commands with a developer’s account.