The ReadJuly 23, 2026via InfoQ AI/ML
Indirect Prompt Injection Exploits GitHub's AI Agent to Leak Private Repository Data
Why it matters
Prompt injection vulnerabilities in agentic AI systems represent a critical security governance issue for enterprises deploying autonomous agents at scale. This real-world exploit demonstrates that AI safety considerations must extend beyond model training to production workflow architecture.
Key signals
- Exploit name: GitLost
- Vector: Indirect prompt injection via public GitHub issues
- Target: GitHub Agentic Workflows
- Impact: Private repository data leakage to public comments
- Discovery source: Noma Security
- Attack method: Concealed instructions embedded in public issues bypass AI agent security safeguards
- Published: July 23, 2026
The hook
GitHub's AI agents just became a vector for stealing private code. Here's how attackers exploited prompt injection to leak confidential repos.
GitLost is a prompt-injection exploit discovered by Noma Security that tricks GitHub's new Agentic Workflows into leaking private data. By embedding concealed instructions within public GitHub issues, attackers can circumvent security safeguards and induce AI agents to reveal confidential information in public comments.
By Sergio De Simone