Spammers are using nearly invisible Unicode tags to break the text patterns that email filters recognize while leaving messages readable to people. The technique, often called ASCII smuggling, previously drew attention because attackers used it to conceal prompt-injection instructions inside content processed by AI agents.
A block of 128 Unicode tag characters mirrors part of ordinary ASCII but is not normally visible on screen. Inserting one inside a word such as “funding” can make a filter or tokenizer process separate fragments, even though the recipient sees the complete word. Campaigns have used the trick to obscure finance terms, dollar amounts and other signals associated with bulk mail.
Microsoft Defender for Office recorded a sharp increase earlier this year. Daily signatures jumped from roughly 21,000 to more than 1.3 million on one day, then reached 2.5 million within four days. The wave continued for months before dropping sharply in mid-May.
Hidden characters are not a new spam tactic; zero-width and non-breaking spaces have served similar purposes for years. The newer tag block matters because modern filters increasingly rely on machine-learning models and subword tokenization. Defenses need to normalize or explicitly detect these characters before classification, while preserving evidence for investigation. Visual rendering with optical character recognition can provide another view, but it adds processing cost.