CamoLeak: How Hackers Exploit GitHub Copilot’s Invisible Markdown to Steal Your API Keys and Cloud Secrets (CVE-2025-59145) + Video

Introduction

AI‑powered coding assistants like GitHub Copilot have become indispensable, but their deep integration with private repositories creates a dangerous attack surface. The recently disclosed CamoLeak vulnerability (CVE‑2025‑59145, CVSS 9.6) demonstrates how attackers can weaponize Copilot’s trusted access—hiding malicious instructions inside GitHub’s invisible markdown comment syntax to exfiltrate source code, API keys, and cloud secrets without executing a single line of malicious code.

Learning Objectives

• Understand the mechanics of CVE‑2025‑59145 (CamoLeak) and how invisible markdown syntax enables data theft.
• Learn detection techniques to identify hidden injection attacks in repositories and AI prompts.
• Implement practical mitigation strategies, including network monitoring, secrets scanning, and Copilot hardening.

You Should Know

1. The CamoLeak Attack Vector: Weaponizing Trusted AI Context

CamoLeak abuses the fact that GitHub Copilot Chat has access to the entire conversation context, including markdown comments that are rendered invisible to human reviewers. Attackers craft a pull request or issue comment containing zero‑width joiners, invisible Unicode characters, or HTML comments (e.g., `