Overview
- Microsoft said the campaign that began in early February 2026 surged to millions of detected messages per day and then fell sharply after May 15, 2026, with Defender telemetry showing weekday volumes in the low‑millions at the peak.
- Attackers placed characters from the Unicode Tags block (U+E0000–U+E007F) inside words so the text looks normal to people but breaks literal keyword and regex matches used by filters; one example turns “funding” into a visible “funding” while the underlying bytes include a tag code point.
- The operation scaled using hundreds of disposable finance‑themed domains and marketing automation routed through ActiveCampaign click‑tracking domains, a setup that can defeat reputation‑based filtering when messages come from a reputable platform.
- Defenders face technical limits because common Unicode normalization does not reliably remove tag characters and different tokenizers handle them differently, so Microsoft and researchers advise stripping or comparing tag‑stripped text and testing the exact tokenizers used by downstream AI systems.
- The shift shows an AI‑era evasion technique—known as ASCII smuggling—moving into mass phishing, which raises risks both for email filters and for systems that ingest external text and for the businesses and people targeted by credential‑harvest and loan‑scam lures.