Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

WeaknessASC-2026-0039Other weakness

Command allowlist and approval bypass in coding agents

MediumMitigated

Agents decide whether a command is safe by parsing it. Shell features (pipes, flags, substitutions, directory changes) repeatedly slip past those checks and run unapproved actions.

How to detect it

  • Agent versions with known approval-bypass CVEs (Claude Code has had several)
  • Approved commands that accept arbitrary arguments

How to resolve it

  • Update coding agents promptly; most bypasses are fixed within days
  • Do not allowlist commands that can run arbitrary code (find -exec, sed -e, git with hooks)
  • Rely on the sandbox, not the parser, as the boundary

How to investigate

  1. Check the agent version against advisories
  2. Review the commands approved in recent sessions

Sources

Open it in the Learning Center's vulnerability database