In the wild
Realized harm
MaliciousCorgi: AI coding-assistant extensions silently exfiltrating source from 1.5 million developers
AIR-2026-0015 AIR record · 2026-01-koi-maliciouscorgi-ai-extensions
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
The moment you open any file - not interact with it, just open it - the extension reads its entire contents, encodes it as Base64, and sends it to a webview containing a hidden tracking iframe.koi.ai
The record
Disclosed
2026-01-22
Vendor
third-party publishers
Product
VS Code extensions 'ChatGPT - Chinese version' (WhenSunset) and 'ChatMoss / CodeMoss' (zhukunpeng)
Risk domain
Tool misuse › Supply chain vulnerabilities
Also touches
Credential theft, Data exfiltration channels
System type
Generative model
Who acted
AI was the target (contested) — malicious IDE extensions harvested source; the exfil code is not the AI
Attack surface
Skill / plugin
Autonomy level
Assistant
Initial vector
Malicious marketplace package
Impact type
Source code exfiltration, Behavioral profiling, Data exfiltration, Credential theft, Ip theft
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI04 – Agentic Supply Chain Vulnerabilities
- MITRE ATLAS
- AML.T0040 – AI Supply Chain Compromise
- NIST AI RMF
- MAP 4.1
- EU AI Act
- Annex III §2
- ISO/IEC
- TR 24028; 42001; 23894
- AIUC-1
- B. Security
- Scope
- tool failure
- Mitigation
- Scan dependencies and container images for vulnerabilities; Use trusted sources.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
1.35 million and 150,000 installs respectively; both still live at publication
Capability abused
Defense & remediation
Recorded defense
Microsoft VS Code Marketplace review
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
None recorded; marketplace response not addressed in the post
Taxonomy & references
Curator notes
Author Tuval Admoni (Koi). Three exfiltration channels: real-time file monitoring, server-commanded mass harvest of up to 50 files, and a four-SDK profiling engine. impact_realized true because active exfiltration from installed users is described as ongoing. Quote's em-dashes normalized to hyphens. Also reported as: 'MaliciousCorgi: two AI coding extensions with 1.5M installs exfiltrated developer files from the VS Code Marketplace'.
Supporting sources
Cite as
Enkrypt AI. "MaliciousCorgi: AI coding-assistant extensions silently exfiltrating source from 1.5 million developers". AIR, AIR-2026-0015. https://agentincident.vercel.app/i/AIR-2026-0015/ (accessed 2026-10-19).