REMOTE PROMPT INJECTION IN GITLAB DUO LEADS TO SOURCE CODE THEFT

Summary from the TLDR InfoSec newsletter

GitLab's AI assistant "Duo" contained an indirect prompt injection vulnerability that allows attackers to manipulate it through hidden prompts in code. This could enable source code theft, phishing attacks, malware distribution, and data exfiltration. Duo's deep DevSecOps integration and browser-rendered Markdown responses create an unusually wide attack surface compared to typical AI coding assistants.