Disclosed vulnerability
PromptJacking: command injection in three official Claude Desktop connectors turns prompts into RCE
AIR-2025-0171 AIR record · 2025-11-koi-promptjacking-claude-desktop
From the supporting source
Claude Desktop Extensions? They run fully unsandboxed on your machine, with full system permissions.koi.ai
The record
Disclosed
2025-11-05
Vendor
Anthropic
Product
Claude Desktop Chrome, iMessage and Apple Notes connectors
Risk domain
Access control and permissions › Privilege escalation
Also touches
Confused deputy
System type
Generative model
Who acted
Agent acted — the connectors executed AppleScript built from model-controlled input
Attack surface
Skill / plugin
Autonomy level
Assistant
Initial vector
Indirect prompt injection to command injection
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
Claude Desktop users with the official extensions installed
Capability abused
Defense & remediation
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
Claude Desktop extensions v0.1.9, 2025-08-28 (verified 2025-09-19)
Taxonomy & references
Curator notes
Author Oren Yomtov. Unescaped interpolation of model-controlled input into AppleScript. Three high-severity issues, CVSS 8.9 each; reported 2025-07-03. Koi does not list CVE ids in the post.
Supporting sources
Cite as
Enkrypt AI. "PromptJacking: command injection in three official Claude Desktop connectors turns prompts into RCE". AIR, AIR-2025-0171. https://agentincident.vercel.app/i/AIR-2025-0171/ (accessed 2026-10-19).