RESEARCHERS FLAG FLAW IN GOOGLE'S AI CODING ASSISTANT THAT ALLOWED FOR ‘SILENT' CODE EXFILTRATION

Summary from the TLDR InfoSec newsletter

There was a vulnerability in Google's Gemini CLI that allowed silent code execution and credential theft through prompt injection in context files - Google patched it on July 25 after initially deeming it low severity and later escalating it to Priority One.