A recent Ars Technica article reports that ASCII smuggling, a technique that hides malicious prompts inside a rarely visible Unicode block, is being repurposed by spammers to bypass email filters. The approach leverages the Unicode Tags block (128 characters) that mirrors ASCII nearly exactly but remains largely invisible to readers.
When embedded in messages, these tags can render sophisticated instructions for automated systems while appearing benign to human eyes, helping malicious content dodge simple keyword or string-based detectors.
Microsoft’s analysis describes a sharp rise in this technique’s use in February 2026, with ASCII-smuggling signatures detected by Defender for Office leaping from around 21,000 per day to more than 1.3 million, peaking near 2.5 million within four days. The activity persisted for months before subsiding in mid-May. Spammers embed invisible text within words (for example, inside “funding”) to thwart pattern matching, while still letting recipients see ordinary words.
The tactic also aims to undermine ML- and NLP-based spam and phishing filters, since tokenisers can be tricked if the hidden characters are present. Microsoft notes that filters relying on image or token-based detection may still miss such content unless they account for hidden Unicode, and provides guidance on adapting filters accordingly.
The piece cites ongoing public commentary from Microsoft and links to prior reporting on invisible text as part of the evolving landscape of prompt-injection and phishing evasion.