‘GitLost’: researchers tricked GitHub’s AI agent into leaking private repos

The Next WebPublished Jul 8, 2026

Security researchers at Noma Labs discovered a vulnerability called GitLost in GitHub's AI Agentic Workflows that allows private repository contents to be leaked publicly via a simple issue with crafted prompts. The flaw exploits prompt injection, bypassing GitHub's guardrails, and currently has no code fix or official documentation from GitHub.

#agentic-workflows #github #gitlost #noma-labs #prompt-injection

‘GitLost’: researchers tricked GitHub’s AI agent into leaking private repos

More in Dev

Code Verification In The Agentic Era

Malicious File in Downloads Folder Tries to Trick AI Coding Agent Into Installing Malware

Tsinghua AI Open-Sources AKV, a Cache Engineering Server for LLM Agents on SGLang