GuardFall Flaw Exposes 10 of 11 Open Source AI Agents to Supply Chain Attacks

Adversa AI has identified a structural security flaw, dubbed GuardFall, affecting ten out of eleven popular open-source AI coding agents. This flaw allows attackers to exploit shell injection vulnerabilities, bypassing command filters through outdated Bash tricks. The affected agents include Hermes, OpenCode, and Roo-code, which run with full developer account authority, posing significant supply chain risks. Only one agent was found to be secure against this vulnerability. The flaw stems from a failure to guard against legacy Bash shell tricks such as quote removal and $IFS spacing. The findings were published in a report titled 'GuardFall: a universal shell injection vulnerability in open-source AI agents.' This situation highlights the ongoing security challenges posed by long-standing coding practices in modern AI tools.

Key Points: • Ten out of eleven popular open-source AI agents are vulnerable to shell injection attacks. • The flaw, named GuardFall, allows attackers to bypass command filters using legacy Bash tricks. • Only one agent tested was found to be secure against this vulnerability.