AI-Generated Copilot Code Led to Snowflake Jira Compromise
TL;DR. An AI-generated fix by GitHub Copilot introduced a critical vulnerability in a Snowflake repository, allowing unauthenticated access to Jira. - Wiz Research's AI-powered "Red Agent" autonomously discovered the vulnerability five days after its introduction. - The flaw enabled script injection via crafted GitHub issue titles, granting access to Snowflake's Jira portal. - Snowflake remediated the issue quickly and confirmed no data exfiltration beyond Wiz's proof-of-concept.
- GitHub Copilot's 'Autofix' feature introduced a critical workflow injection vulnerability in a public Snowflake repository.
- The vulnerability allowed unauthenticated users to execute arbitrary commands and access Snowflake's Jira portal.
- Wiz Research's 'Red Agent,' an AI-powered security tool, autonomously discovered and reported the flaw within days.
- Snowflake promptly remediated the issue and verified that only Wiz accessed data during testing.
Sources
- AI-Generated GitHub Copilot "Autofix" Allowed Compromise of Snowflake's Jira — wiz.io
- theregister.com — theregister.com