// Security Advisory Triage

PraisonAI sandbox bypass allows unrestricted command execution

GitHub advisories — pip · published 2026-10-09 13:20 UTC
ACTION NEEDED PATCH_NOW SECURITYAPI
WHAT CHANGEDCVE-2026-60085 published 8 October 2026. PraisonAI versions before 4.6.78 do not enforce SecurityPolicy restrictions on commands, paths, or imports when using the default SubprocessSandbox backend. An attacker can execute arbitrary commands despite configured security policies.
WHY IT MATTERSAgent runtimes that rely on SecurityPolicy to isolate tool execution or constrain prompt-driven actions are not actually isolated. Any code path that invokes tools through the sandbox can be exploited to run commands the policy was meant to block.
WHAT TO DOUpgrade praisonai to 4.6.78 or later. If you cannot upgrade immediately, audit every agent configuration that uses SubprocessSandbox and either disable tool execution or switch to a different isolation mechanism until patched.
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-5r6c-gj4g-r697
+summary: PraisonAI: SecurityPolicy command/path/import restrictions are completely unenforced by the default SubprocessSandbox backend
+severity: high
+published_at: 2026-10-08T21:58:11Z
+cve_id: CVE-2026-60085
+vulnerabilities[].package.name: praisonai
+vulnerabilities[].first_patched_version: 4.6.78
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 reasoningPraisonAI is an AI agent framework; sandbox bypass in default backend affects security posture of services using it.
signalsAI framework package (praisonai) in dependency tree · sandbox/security policy bypass in agent runtime · high severity · new CVE with patched version available · affects prompt/tool execution isolation
suppression—
triage modelanthropic/claude-haiku-4.5 ($0.00000)
writer modelanthropic/claude-sonnet-4.5 ($0.00000)