Spammers Hijack AI's Invisible Text Trick

Spammers are now employing a sophisticated technique called ASCII smuggling, previously used to subvert AI agents, to evade email filters. This method involves embedding hidden instructions within emails using a special range of invisible unicode tags that AI can read but humans cannot.

This tactic allows spammers to circumvent keyword-based filters that commonly flag terms like "credit" or "term." By inserting these invisible characters, they can disguise words that would normally trigger a spam alert. Microsoft observed a massive spike in detected ASCII smuggling signatures starting in early February, with detections soaring from thousands to millions daily.

The effectiveness of ASCII smuggling lies in its ability to fool both simple keyword matching and more advanced machine learning and natural language processing models used in spam detection. Unless filters can visually analyze message content like a human, they may miss these disguised malicious prompts. Microsoft has since provided guidance for improving filter defenses against this evolving threat.