Snowflake's GitHub Actions Injection: How an AI Red-Team Agent Found What Copilot Missed

How Wiz's autonomous AI red-team agent found a GitHub Actions command injection bug in a public Snowflake repository that Copilot-assisted review had missed.

Snowflake's GitHub Actions Injection: How an AI Red-Team Agent Found What Copilot Missed
Photo by Ilya Pavlov / Unsplash

Researchers at Wiz used an autonomous AI security agent to uncover a command injection vulnerability in a public Snowflake GitHub repository just five days after the vulnerable code shipped — a bug that had already passed automated review from GitHub Copilot.

Vulnerability Intelligence: Repository: snowflakedb/snowflake-connector-net (public) Vulnerability Type: GitHub Actions workflow command injection Severity: Critical (unauthenticated, pre-auth command execution in CI runner) Status: Patched same day as disclosure Discovered By: Wiz Red Agent (autonomous AI security research tool)

What Happened

Wiz's autonomous AI red-team agent identified a GitHub Actions workflow injection vulnerability in jira_issue.yml, a workflow in Snowflake's public snowflake-connector-net repository designed to automatically create Jira tickets whenever someone opened a GitHub issue. A pull request merged on June 18, 2026 had replaced a safer implementation — one that passed values through environment variables and jq — with direct interpolation of the untrusted issue title into a shell script. Wiz's agent found the bug five days later and reported it to Snowflake via HackerOne on June 23, 2026; Snowflake merged a fix that same day.

Who's Affected

The vulnerability lived in a public, unauthenticated code path: anyone able to open a GitHub issue on the repository — which, being public, means essentially anyone with a GitHub account — could have crafted an issue title that executed arbitrary commands inside the workflow's CI runner. That runner's access to repository secrets and internal tooling made this a meaningful supply-chain risk, not just a nuisance.

Technical Root Cause

GitHub Actions workflows that interpolate untrusted input (like an issue title) directly into run steps via ${{ }} expression syntax are a well-known injection class — GitHub's own documentation warns against it — but the pattern still recurs because it's easy to introduce during a routine refactor, as happened here when a safer environment-variable-based implementation was swapped out. Notably, the vulnerable pull request had passed review that included GitHub Copilot, underscoring that AI code review isn't a substitute for dedicated security tooling.

Remediation

Audit GitHub Actions workflows for any run step that interpolates issue or PR titles and other user-controlled context values directly into shell commands; pass such values through environment variables instead and reference them safely inside the script. Treat GitHub issue titles, PR titles, and other user-controlled workflow inputs as untrusted by default. Organizations running public repositories with automation triggered by issues or PRs should specifically test for this injection class rather than relying solely on AI-assisted code review to catch it.

This is part of ColibriSec's weekly security roundup for the week of August 20, 2026.

Read more

Guía de Defensa para CISOs 6 de Noviembre de 2026 Xen XSA-523 Alpine OpenSSH sshhole LXD mTLS cPGuard y Cadena de Frío OT

Guía de Defensa para CISOs del 6 de Noviembre de 2026: Blindaje de Xen Dom0, OpenSSH musl en Alpine, LXD mTLS, Paneles cPGuard/CWP, Controladores OT de Cadena de Frío y AD CS

📌Security Roundup Series: Semana del 6 de Noviembre de 2026 • 9 min de lectura deep dive🚨Vulnerability Intelligence: CVE ID: CVE-2026-119977, CVE-2026-107441, XSA-523, CVE-2026-112816, CVE-2026-123158 y CVE-2026-58868 (Grant Copy en Hipervisor, Validación mTLS, Retorno de Error en musl crypt(a.plerdy.com), TOCTOU y Control de Acceso OT) Severity: MANUAL

By James Luther
Campaña Poisoned Source de APT29 en Plerdy, Kali365 y VoidProxy PhaaS, Teams AD CS y Brechas en Queens Library Myriad y mSecure

Campaña "Poisoned Source" de APT29 (Plerdy), PhaaS con Señuelos de IA (Kali365 y VoidProxy), Toma de Control AD CS vía Teams y Brechas Globales (Queens Library, Myriad Genetics, mSecure y Japón)

📌Security Roundup Series: Semana del 6 de Noviembre de 2026 • 9 min de lectura deep dive🏛️Incident Overview: Target / Organization: Sitios con Plerdy SEO (APT29), Empresas M365/Google (Kali365/VoidProxy), Queens Public Library, Myriad Genetics y mSecure Threat Actor / Attribution: APT29 (Midnight Blizzard), Operadores PhaaS Kali365 / VoidProxy, Brain Cipher Ransomware

By James Luther