Attackers Hide AI Prompt Injections Inside Phishing Emails

Barracuda analyzed a phishing campaign that embeds hidden prompt-injection instructions inside otherwise legitimate-appearing emails to target both human recipients and AI assistants that summarize inboxes; attackers used techniques such as HTML comments, invisible CSS text, Base64 encoding and zero-width characters to hide instructions that can cause credential theft, fraudulent wire transfers, data leaks or malicious code changes, and the vendor recommends stripping hidden elements, detecting instruction-override language, sandboxing AI, validating outputs and requiring human approval for critical actions.