Attackers conceal phishing lures using invisible Unicode characters

Attackers have adopted the ASCII smuggling technique in phishing campaigns by embedding invisible Unicode characters (from the Unicode Tags block U+E0000–U+E007F) within finance-related lure words to evade email security filters that rely on keyword detection. Microsoft threat researchers identified a large-scale campaign delivering up to 2.37 million phishing emails daily in February 2026, which gradually declined but remained active as of May 2026. The campaign used sender domains with finance-themed keywords and leveraged the ActiveCampaign email-marketing platform infrastructure. Despite the evasion technique, Microsoft Defender for Office 365 detected over 99% of these messages using other indicators such as sender reputation and domain checks. Microsoft advises defenders to strip or normalize invisible Unicode characters before keyword or signature-based detection and to treat unexpected tag-block characters as anomalies to improve detection and mitigate related AI prompt-injection risks.

The phishing campaign successfully evades keyword-based email filters by using invisible Unicode characters, potentially increasing the likelihood of phishing emails reaching users' inboxes. However, Microsoft Defender still detects the vast majority (over 99%) of these messages using other detection signals. The campaign targets finance-related themes such as funding, loans, and credit services and abuses a legitimate email marketing platform. The evasion technique complicates detection but does not fully bypass existing security controls.

Microsoft recommends stripping or normalizing invisible Unicode tag characters and other invisible code points before applying keyword, regex, or signature-based detection in email security solutions. Defenders should treat unexpected Unicode tag-block characters as strong anomalies. Applying the same normalization before passing email content to AI assistants can mitigate prompt-injection attack risks. ActiveCampaign has stated that its moderation systems detect invisible Unicode characters similarly to unobfuscated text and treat heavy use as suspicious. No official patch is applicable as this is an evasion technique rather than a software vulnerability.