ทันเอไอ

Wiz reveals AI detected Snowflake CI/CD vulnerability after Copilot patch enabled Jira API key theft

Wiz reported that Red Agent, an AI security testing tool, discovered a command injection vulnerability in Snowflake's GitHub Actions caused by a newly modified CI/CD script, successfully stealing Jira API keys before Snowflake quickly patched the flaw and rotated credentials on the same day it was notified.

Cybersecurity firm Wiz reported the results of a vulnerability test on GitHub Actions under Snowflake's account, where Red Agent—an AI used by Wiz for security auditing—discovered and exploited a vulnerability to successfully steal Jira API keys.

The vulnerability was found in the `snowflakedb/snowflake-connector-net` repository, where a workflow automatically created a Jira issue whenever a GitHub issue was opened. Originally, the script retrieved values via the jq command and environment variables, which is safer. However, a patch co-authored by Copilot Autofix changed this to execute bash directly from a GitHub variable, resulting in a command injection vulnerability via GitHub Issue titles.

Nevertheless, sources confirmed it remains unclear whether the problematic code was entirely AI-generated. Copilot Autofix merely participated in the pull request and did not detect the vulnerability during review, meaning it should not be directly blamed on the AI.

Snowflake was notified by Wiz on June 23, 2026, and resolved the vulnerability while immediately rotating the credentials on the same day. Investigations confirmed that no external parties other than Wiz researchers accessed the data during that period. g This case serves as another example highlighting the risks of modern CI/CD pipelines, where minor code modifications can turn into severe vulnerabilities—especially when user-inputted data, such as issue titles, is executed directly in scripts without validation.