Researchers hide prompt injection inside a PNG image, fooling AI coding assistants into exposing data

Digital Trends reports that security researchers have demonstrated a new attack where a seemingly harmless PNG image contains a hidden prompt injection. The technique successfully fools AI coding assistants into exposing sensitive data.

• Researchers hid a prompt injection inside a PNG, and AI fell for it
• 'Ghostcommit' hides prompt injection in images to fool AI agents, steal secrets