// Security Advisory Triage

AG2 newline injection bypasses SandboxShellTool security controls

GitHub advisories — pip · published 2026-11-14 06:24 UTC
ACTION NEEDED PATCH_NOW SECURITY
WHAT CHANGEDCVE-2026-113663 published 10 November 2026. Newline injection in AG2's SandboxShellTool bypasses allowed=[...] and readonly=True restrictions, enabling arbitrary command execution. Affects ag2 and autogen packages. Patched in version 1.1.1.
WHY IT MATTERSAn attacker can execute arbitrary commands despite configured security controls. The allowed list and readonly flag do not prevent exploitation when input contains newlines.
WHAT TO DOUpdate ag2 or autogen to version 1.1.1 or later. Verify you are running SandboxShellTool before patching.
Model-reported confidence 95% (not measured accuracy) Primary source ↗ written by anthropic/claude-sonnet-4.5
Show the reasoning — raw diff, triage verdict, confidence
RAW DIFF
+ghsa_id: GHSA-2h4w-29cv-2cmv
+summary: AG2: Newline injection in SandboxShellTool bypasses allowed=[...] and readonly=True, enabling arbitrary command execution
+severity: high
+published_at: 2026-11-10T15:57:17Z
+cve_id: CVE-2026-113663
+vulnerabilities[].package.name: ag2, autogen
+vulnerabilities[].first_patched_version: 1.1.1,
HOW IT WAS JUDGED
Triage signals below are model-reported assertions, preserved verbatim; they are not independent arithmetic verification.
triage verdictMATERIAL at 0.92, needed 0.5
triage reasoningPython agent framework with RCE via security control bypass. Platform leads run AI agents; this is actionable.
signalsknown exploitation vector: newline injection bypasses security controls · arbitrary command execution capability · high severity rating · affects Python package in dependency tree (ag2/autogen) · patched version available (1.1.1)
suppression—
triage modelanthropic/claude-haiku-4.5 ($0.00000)
writer modelanthropic/claude-sonnet-4.5 ($0.00000)