Disclosed vulnerability
Authorization bypass in claude-code-action lets any GitHub App-authored issue drive prompt injection into CI
AIR-2026-0126 AIR record · 2026-06-csa-claude-code-action-bypass
From the supporting source
The function unconditionally permitted any actor whose identity string ended in `[bot]`, on the assumption that bot actors would be legitimate GitHub Apps installed by repository owners. This assumption was incorrect.labs.cloudsecurityalliance.org
The record
Disclosed
2026-06-01
Vendor
Anthropic
Product
claude-code-action (< v1.0.94) and @anthropic-ai/claude-code (< v1.0.93)
Risk domain
Tool misuse › Supply chain vulnerabilities
Also touches
Confused deputy
System type
Generative model
Who acted
Agent acted — the CI agent acted on the injected issue with repository write access
Attack surface
Coding agent
Autonomy level
Semi-autonomous
Initial vector
Indirect prompt injection issue
Impact type
Repository compromise, Supply chain backdoor
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
Public repositories running claude-code-action; any GitHub App has implicit read access and can open issues
Capability abused
Defense & remediation
Recorded defense
checkWritePermissions actor authorization check
Guardrail kinds
Permission scope
Guardrail outcome
Silent
Fix shipped
claude-code-action v1.0.94 and @anthropic-ai/claude-code v1.0.93; fix deployed within four days of report
Taxonomy & references
Curator notes
CSA AI Safety Initiative research note (published 2026-06-05) writing up RyotaK of GMO Flatt Security's finding. Reported 2026-01-12, fixed in four days, publicly disclosed 2026-06-01. Alias GHSA-xq4m-mc3c-vvg3. The record is dated by public disclosure; the CSA note is used as the primary source of record because it is the writeup available in citable form.
Supporting sources
Cite as
Enkrypt AI. "Authorization bypass in claude-code-action lets any GitHub App-authored issue drive prompt injection into CI". AIR, AIR-2026-0126. https://agentincident.vercel.app/i/AIR-2026-0126/ (accessed 2026-10-19).