Phishers Evade Filters with Invisible Unicode Characters

A phishing campaign has been using invisible Unicode tag characters to evade email filters. The attackers repurposed ASCII smuggling techniques, popularized in AI prompt injection research, to hide financial lure words from recipients while still allowing them to be processed by software and AI models.

Microsoft researchers detected the campaign after a surge in signature hits on their Defender for Office 365 prompt injection protection research. The number of messages jumped from about 21,000 to over 1.3 million in one day, with some weekdays reaching 2.37 million. Over 99% of these messages were flagged by other Defender protections.

The attackers used a specific block of Unicode tag characters that have recently drawn attention in AI security research. This technique could expose a blind spot in email security systems if they don't properly handle Unicode tag characters before analyzing message content. Microsoft recommends stripping or normalizing invisible Unicode characters before applying spam and phishing content signatures.