‘GitLost’ vulnerability let GitHub’s AI workflows leak private repositories
A single crafted GitHub issue just exposed private repositories. Here's why AI workflow security is now a board-level risk.
Why it matters
GitLost exposes a critical gap in AI agent security: prompt injection vulnerabilities in production workflows can bypass authentication controls and leak sensitive data at scale. This is the first major security incident in GitHub's agentic features, signaling that AI-native infrastructure needs fundamentally different threat modeling than traditional apps.
The key facts
7 to knowVulnerability: Prompt injection in GitHub Agentic Workflows feature
Impact: Unauthenticated attacker can access private code repositories
Attack vector: Single crafted issue posted in public repository
Discovered by: Noma Security Inc. (AI security company)
Vulnerability name: GitLost
Affected product: GitHub Agentic Workflows (AI feature)
Authentication bypass: No credentials required for exploitation
Go to the source
SiliconAnglesiliconangle.com
Publisher excerpt: Researchers at artificial intelligence security company Noma Security Inc. today disclosed a critical prompt injection vulnerability in GitHub Inc.’s new Agentic Workflows feature that allowed an unauthenticated attacker to siphon data from private code repositories by posting a single crafted…