AI-Generated GitHub Copilot “Autofix” Allowed Compromise of Snowflake's Jira
Points and comments are a snapshot, not live.
Wiz Red Agent exploited an AI-generated GitHub Copilot Autofix that introduced a shell injection vulnerability in Snowflake's Jira.
Wiz Research's autonomous 'Red Agent' discovered a script injection vulnerability in Snowflake's public snowflake-connector-net repository via HackerOne. The flaw, introduced by PR #1218 (merged June 18, 2026), allowed unauthenticated users to execute arbitrary commands on GitHub Actions runners via a crafted issue title. GitHub Advanced Security did not flag the injection. Wiz exploited it to exfiltrate a Jira token granting read access across Snowflake's engineering and security projects. Snowflake patched the same day (June 23) and rotated the credential. The vulnerability was live only five days before automated discovery.
What commenters are saying
Commenters were sharply critical of the reliance on AI-generated code and automated reviews, arguing that human peer review remains essential and that testing tools like actionlint or zizmor could have caught the quote injection. Two camps emerged: those decrying the reuse of insecure shell patterns and the naive if: condition, and those noting the risk of auto-approving 'minor' AI PRs. A sub-thread debated whether gatekeeping feature requests is justified, with some predicting that lowered AI code-generation costs will overwhelm that model.