Research demo
Prompt injection in a GitHub issue makes GitHub Copilot Agent plant a backdoor in a lock file
AIR-2025-0085 AIR record · 2025-08-trailofbits-copilot-backdoor-injection
From the supporting source
The issue contains a prompt injection attack that causes Copilot to discreetly insert a backdoor for the attacker in its pull request, which the victim merges.blog.trailofbits.com
The record
Disclosed
2025-08-06
Vendor
GitHub / Microsoft
Product
GitHub Copilot Agent
Risk domain
Access control and permissions › Privilege escalation
Also touches
Supply chain vulnerabilities, Confused deputy
System type
Generative model
Who acted
Agent acted — the Copilot agent wrote a backdoored lock file and opened the PR
Attack surface
Coding agent
Autonomy level
Semi-autonomous
Initial vector
Indirect prompt injection issue
Impact type
Supply chain backdoor, 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
Public repositories where Copilot Agent acts on issues
Capability abused
Defense & remediation
Recorded defense
GitHub markdown sanitization stripping <picture> tags
Guardrail kinds
Output filter
Guardrail outcome
Silent
Taxonomy & references
Curator notes
Author Kevin Higgs. No CVE/GHSA; framed as attacker-side prompt-injection engineering rather than a coordinated disclosure.
Supporting sources
Cite as
Enkrypt AI. "Prompt injection in a GitHub issue makes GitHub Copilot Agent plant a backdoor in a lock file". AIR, AIR-2025-0085. https://agentincident.vercel.app/i/AIR-2025-0085/ (accessed 2026-10-19).