GitHub Copilot just wrote the code that handed an attacker Snowflake's entire Jira. An AI agent found it in 5 days.

An AI coding assistant co-authored a pull request that gutted Snowflake's own security.

Copilot Autofix merged PR #1218 on June 18.
It replaced a safe input pattern with raw string interpolation.
The commit message credited "Copilot Autofix powered by AI."

Five days later, Wiz's autonomous security agent — Red Agent — found it.
Exploited it. Exfiltrated Snowflake's Jira credentials.
Engineering. Security compliance. Bug bounty tracking. All accessible.

The safe pattern existed for a reason.
It used env variables and jq to sanitize untrusted input.
Copilot removed it. Replaced it with a shell command that trusted GitHub issue titles as code.

The "security gate" that was supposed to catch this?
It checked pull_request.user.login on an issues event.
Pull request is always null on issues events.
The gate was always true. Every user passed.

This is the enterprise AI security crisis in one incident.

AI tools writing code without understanding why patterns exist.
Automated agents discovering and weaponizing those flaws in days.
Governance controls that look protective but are structurally broken.

Your CI/CD pipeline has the same exposure.
Every repo using AI coding assistants has the same exposure.

Audit every AI-generated commit in your repos today.
Check whether your security gates actually evaluate the events they claim to protect.
If you cannot answer those questions in five minutes, you are already exposed.