Agent Incident RegistryAIR
v1 · updated 19 Oct 2026
Disclosed vulnerability

CVE-2026-85169: n8n $fromAI prototype leak lets an expression escape the sandbox and run code on the host

AIR-2026-0258 AIR record · 2026-09-n8n-fromai-sandbox-escape-rce
From the supporting source
compile code running in the main n8n process, from workflow-build privilege alone.github.com
The record
Disclosed
2026-09-03
Vendor
n8n
Product
n8n
Risk domain
Access control and permissions › Privilege escalation
System type
Generative model
Who acted
AI was the target (contested) — expression sandbox escape in an automation platform; contested because $fromAI is the agent-parameter mechanism
Attack surface
Agent framework
Autonomy level
Assistant
Initial vector
Authenticated workflow author expression
Impact type
Remote code execution
Confidence
High
Submitted by
EAI Internal
Crosswalk
OWASP Agentic Top 10 (2026)
ASI03 – Identity and Privilege Abuse
MITRE ATLAS
AML.T0055 – Unsecured Credentials
NIST AI RMF
GOVERN 6.1
EU AI Act
Article 13
ISO/IEC
TR 24028; 42001; 23894
AIUC-1
B. Security
Scope
tool misuse
Mitigation
Apply least-privilege principle; Regular policy audits.
Taxonomy
Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
n8n before 1.123.73, 2.35.4 and 2.36.2 for users able to build workflows
Capability abused
fromAI_placeholder_resolutionexpression_evaluation
Defense & remediation
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
n8n 1.123.73, 2.35.4 and 2.36.2
Taxonomy & references
OWASP ASI05 Unexpected Code Execution (RCE)CVE-2026-85169sandbox_escapeprototype_chainai_expression_helper
Curator notes
GitHub advisory GHSA-9x83-43r8-5hwc was published 2026-08-19 (CVSS 4.0 score 8.7); the CVE record was published 2026-09-03, which is the date used here. The $fromAI helper, which resolves placeholders that a model fills in, returned a live host-prototype reference, letting an expression reach the Function constructor and run code in the main n8n process. Requires workflow-build privilege. The advisory is a vulnerability description; no exploitation is reported.
Cite as
Enkrypt AI. "CVE-2026-85169: n8n $fromAI prototype leak lets an expression escape the sandbox and run code on the host". AIR, AIR-2026-0258. https://agentincident.vercel.app/i/AIR-2026-0258/ (accessed 2026-10-19).