Disclosed vulnerability
ChatGPT Chat with Code plugin: prompt injection turns private GitHub repos public and steals source
AIR-2023-0013 AIR record · 2023-06-openai-chat-with-code-plugin
From the supporting source
Over the last few months I reached out to OpenAI about these issues many times (at least four) with little to no success or feedback.embracethered.com
The record
Disclosed
2023-06-20
Vendor
OpenAI (third-party plugin)
Product
ChatGPT Chat with Code plugin
Risk domain
Agent behaviour › Unsafe actuation
Also touches
Sensitive data exposure, Confused deputy
System type
Generative model
Who acted
Agent acted — the plugin performed OAuth-authorised repo writes, making private repos public
Attack surface
Skill / plugin
Autonomy level
Assistant
Initial vector
Indirect prompt injection web content
Impact type
Unauthorized action, Source code theft, Privacy violation
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI01 – Agent Goal Hijack
- MITRE ATLAS
- AML.T0048 – External Harms
- NIST AI RMF
- MEASURE 2.6; MANAGE 1.3
- EU AI Act
- Annex III §1(c)
- ISO/IEC
- TR 24028; 24029-1; 23894
- AIUC-1
- C. Safety
- Scope
- agent misuse
- Mitigation
- Require confirmation for destructive actions; Implement dry-run modes; Use layered safety controls.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
users who installed the plugin and authorized GitHub OAuth
Capability abused
Defense & remediation
Recorded defense
none - the plugin took actions without user confirmation
Guardrail kinds
None
Guardrail outcome
Not applicable
Fix shipped
the plugin was later removed from the store (see the 2023-07-06 follow-up post); OpenAI added store criteria requiring user confirmation before actions
Taxonomy & references
Curator notes
Johann Rehberger. Earliest widely cited case of an LLM plugin performing destructive OAuth-authorized actions from injected instructions.
Cite as
Enkrypt AI. "ChatGPT Chat with Code plugin: prompt injection turns private GitHub repos public and steals source". AIR, AIR-2023-0013. https://agentincident.vercel.app/i/AIR-2023-0013/ (accessed 2026-10-19).