Guard Fall Exposes Open-Source AI Coding Agents to Decades-Old Shell Injection Risks

Summary: Adversa AI's research, named GuardFall, found a shell-injection bypass that worked against ten of the eleven open-source coding and computer-use agents tested, with only Continue defending against it. The bypass exploits the gap between blocklist filters that read commands as plain text and bash, which strips quotes and expands shortcuts before running, so a filter watching for rm does not flag r''m even though bash executes rm.
Why it matters: These agents run shell commands with your full account access, so a booby-trapped repository or package can reach SSH keys and cloud credentials, and the ten tools that left the gap open carried roughly 548,000 GitHub stars as of May 2026 with no single CVE to track or patch.

Read the original at The Hacker News ↗