Agent Incident RegistryAIR
v1 · updated 19 Oct 2026
Disclosed vulnerability

Bypassing command allowlists turns prompt injection into RCE in three AI coding agents

AIR-2025-0154 AIR record · 2025-10-trailofbits-prompt-injection-rce
From the supporting source
Pre-approved commands create a security drawback: they expose an argument injection attack surface when user input can influence command parameters.blog.trailofbits.com
The record
Disclosed
2025-10-22
Vendor
multiple
Product
Three unnamed AI coding agents (prior work cites Amazon Q, Claude Code, Cursor)
Risk domain
Access control and permissions › Privilege escalation
Also touches
Confused deputy
System type
Generative model
Who acted
Agent acted — the agents executed commands past their allowlists
Attack surface
Coding agent
Autonomy level
Copilot with approval
Initial vector
Indirect prompt injection repository content
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
Agents relying on pre-approved command allowlists without sandboxing
Capability abused
shell_command_executionpre_approved_command_allowlist
Defense & remediation
Recorded defense
Human approval prompts and pre-approved command allowlists
Guardrail kinds
Human approval, Permission scope
Guardrail outcome
Silent
Taxonomy & references
OWASP ASI01 Agent Goal HijackOWASP ASI05 Unexpected Code Execution (RCE)CVE-2025-54795argument_injectionallowlist_bypassrcehuman_approval_bypass
Curator notes
Author Will Vandevanter. Products unnamed 'due to ongoing coordinated disclosure'. CVE-2025-54795 (Claude Code) and GHSA-534m-3w6r-8pqr (Cursor) are cited as prior work, not as this post's own findings.
Cite as
Enkrypt AI. "Bypassing command allowlists turns prompt injection into RCE in three AI coding agents". AIR, AIR-2025-0154. https://agentincident.vercel.app/i/AIR-2025-0154/ (accessed 2026-10-19).