Wiz Red Agent accessed Snowflake Jira 5 days after Copilot-checked flaw went live

Wiz's autonomous AI agent Red Agent discovered and exploited a GitHub Actions script injection vulnerability in Snowflake's CI/CD pipeline, exfiltrating internal Jira access credentials just five days after the flaw was merged in a pull request co-authored by GitHub Copilot and cleared by GitHub Advanced Security. The exploit allowed an unauthenticated attacker to execute arbitrary code in the GitHub Actions runner, highlighting that AI-assisted code review can miss critical security flaws.

Hacker News https://www.wiz.io/blog/red-agent-snowflake-copilot-cicd-bug Wiz Red Agent accessed Snowflake Jira 5 days after Copilot-checked flaw went live Which summary reads better? Pick one — models revealed after.Both summaries are AI-generated. GitHub Copilot Autofix approved a PR that introduced a command-injection flaw in Snowflake’s CI/CD pipeline, letting an unauthenticated attacker execute arbitrary code in their GitHub Actions runner. This means AI-assisted PRs can silently add critical security holes that bypass GitHub Advanced Security scans, so every merged PR—even those marked “safe” by Copilot—now needs a second, manual shell-injection review before deployment. An autonomous AI agent discovered and exploited a GitHub Actions script injection vulnerability to exfiltrate internal Jira access credentials just five days after the flaw was merged in a pull request co-authored by GitHub Copilot and cleared by GitHub Advanced Security. This milestone demonstrates that AI agents can now execute end-to-end exploit chains in the wild faster than typical human patch cycles can secure AI-assisted code. If you are shipping with LLM-based coding agents or autofixers, you must immediately enforce hard boundary isolation and least-privilege token scoping on your CI/CD runners, as traditional static scanners and AI reviewers are actively failing to detect these injection vectors.