Wiz AI Agent Exploits Snowflake GitHub Actions Flaw

A Snowflake open-source workflow used GitHub Actions to automate issue handling but failed to safely process user-controlled issue titles, creating a command-injection vulnerability. An attacker could insert commands into the automation environment and potentially obtain credentials for Snowflake’s internal Jira system. The flaw highlights the security risks that arise when public collaboration tools are connected to privileged corporate services without strict input handling and access controls.

Cybersecurity firm Wiz said its autonomous AI security agent independently debugged the exploit and completed a successful validation attack within five days of the vulnerable workflow going live. The test showed that a crafted GitHub issue title could expose Jira access credentials. Wiz reported the vulnerability to Snowflake, which deployed a fix on the same day it received the disclosure. No evidence of malicious exploitation has been reported.