Unicode Invisible Characters Repurposed From AI Prompt Injection to Phishing Evasion

Microsoft's threat research team discovered in early 2026 that a detection signature designed to catch AI prompt injection attacks began triggering on standard phishing emails. The technique, known as ASCII smuggling, involves hiding text within Unicode tag characters (U+E0000–U+E007F) that are invisible to human readers but readable by parsers. Originally exploited to slip hidden instructions past humans and into AI models, phishing actors adapted the same trick starting around February 9, 2026, using invisible characters to break up filter-triggering keywords like 'funding' so email security tools fail to match them. Because most mail clients strip or ignore these characters during rendering, recipients still see the complete word while keyword-based filters see only a fragmented, unrecognizable string. Microsoft noted the attack volume surged roughly 100-fold, highlighting how an AI-security exploit was repurposed against traditional email defenses without any new malware development.