Wiz Research discovered a critical flaw in GitHub's git infrastructure that enables remote code execution (RCE) through a single `git push` command. The vulnerability (CVE-2026-3854) exploits an injection flaw in GitHub's internal protocol, allowing authenticated attackers to execute arbitrary commands on backend servers. GitHub.com mitigated the issue within 6 hours, but GitHub Enterprise Server (GHES) users face urgent risks - 88% of instances remain vulnerable despite patches being available.

The attack chain leverages unsanitized semicolons in Git push options, which corrupt the X-Stat header used across GitHub's multi-service architecture. By manipulating fields like `rails_env` and `custom_hooks_dir`, attackers can hijack execution paths and bypass security controls. Wiz combined AI-augmented reverse engineering (using IDA MCP) with wire-level analysis to uncover this flaw in GitHub's black-box binaries - a first for closed-source systems.

GitHub fixed GitHub.com but requires GHES administrators to upgrade to version 3.19.3 immediately. The vulnerability affects multiple services including babeld, gitauth, and gitrpcd, with the pre-receive hook binary serving as the final exploitation point. Technical details and mitigation guidance are available in GitHub's security blog.

This incident underscores the evolving threat landscape, where AI-powered tools now enable researchers to audit previously impenetrable systems. As CISO Alexis Wales noted, such findings highlight the need for continuous collaboration between security teams and ethical hackers to address emerging risks in critical infrastructure.