An AI Autofix Broke Snowflake's Security and Exposed Its Jira

An AI code assistant generated a fix that opened a critical security hole in Snowflake's GitHub repository, exposing access to its internal Jira portal.

Security researchers at Wiz discovered a script injection vulnerability in Snowflake's public snowflake-connector-net repository. Five days earlier, a pull request co-authored by GitHub Copilot Autofix removed a safe input-sanitization pattern and replaced it with direct string interpolation in a GitHub Actions workflow. Anyone opening a public GitHub issue could execute arbitrary shell commands on the runner. Wiz used its security tool, Red Agent, to trigger the flaw and extract base64-encoded credentials, granting read access to Snowflake's internal engineering, security, and bug bounty tracking projects on Jira.

Why it matters: Copilot Autofix co-authored the vulnerable change, and GitHub's AI-assisted security review failed to flag it before merging. AI coding assistants often lack context on why safe code patterns exist, making it easy for automated fixes to reintroduce old injection vectors.

Know this: Wiz responsibly disclosed the issue on June 23, 2026. Snowflake patched the workflow the same day, rotated the affected token, and confirmed through audit logs that Wiz was the only actor to touch the endpoint during the five-day exposure window.

AI generated the vulnerability, AI discovered the exploit, and humans patched the workflow before dinner.