Microsoft warns Unicode-tag phishing wave evades filters

Microsoft is warning that a newly observed, finance-themed phishing wave is abusing invisible Unicode tag characters to slip millions of emails past enterprise filters.^[1][2][6] The technique, dubbed ASCII smuggling in recent research, repurposes stealthy Unicode code points originally popularized in AI prompt-injection experiments to evade traditional email security controls.^[1][11]

According to Microsoft Security Research, attackers insert non-rendering Unicode tag characters from the U+E0000–U+E007F block directly into high-signal keywords in subject lines and email bodies.^[1] By splicing characters such as U+E0020 into terms like funding, the campaign breaks keyword and regular-expression matches, so filters no longer recognize the financial lure words even though human recipients see them normally.^[1][6]

Microsoft telemetry ties the Unicode-tag obfuscation to a large-scale finance-themed operation first seen in early February 2026 and still active months later.^[1][11] Reporting on the campaign’s volume indicates peaks of roughly 1 to 2.37 million phishing messages per working day, sent from hundreds of rotating domains but sharing consistent infrastructure patterns.^[2][6][11]

Invisible Unicode characters have surfaced in phishing and spam before, including the use of soft hyphens and word joiners to break up suspicious terms in subject lines, but the latest wave shows a more systematic, large-scale application of the tactic.^[3][5][8] Microsoft previously noted attackers inserting zero-width or non-printing characters to defeat automated analysis, and researchers have since documented similar techniques in JavaScript payload obfuscation and AiTM phishing kits.^[3][9][15]

The Unicode-tag abuse does not hinge on a specific product vulnerability or CVE; instead it exploits how many email gateways and anti-phishing engines tokenize and scan text, aligning with the MITRE ATT&CK “Obfuscated Files or Information” technique T1027.^[1][7] Systems that rely heavily on keyword signatures, simple regular expressions, or unnormalized Unicode input are particularly susceptible, allowing carefully crafted messages to reach inboxes despite other reputation or authentication checks.^[1][5]

To counter the campaign, Microsoft advises defenders to strip or normalize Unicode tag characters in the U+E0000–U+E007F range, as well as other zero-width and invisible code points, before applying spam and phishing signatures.^[1] Security teams should update detection rules to flag abnormal concentrations of non-printing Unicode characters, review email templates and gateways for Unicode normalization gaps, and add user awareness messaging around unexpected finance-related emails that slip through automated filters.^[5][8][12] As Unicode-based obfuscation techniques spread from AI prompt injection research into mainstream phishing, organizations that modernize their content inspection pipelines and threat-hunting playbooks for Unicode edge cases will be better positioned to blunt this emerging class of email evasion.^[1][10][11]

References

1. ASCII smuggling crosses over from AI prompt injection to phishing evasion | Microsoft Security Blog
2. Microsoft Threat Intelligence on X
3. Trend-spotting email techniques: How modern phishing …
5. A phishing with invisible characters in the subject line
6. Unicode-Tag-Smuggling: Neue Phishing-Welle mit Millionen von E-Mails
7. Bypassing Phishing Filters MSc System a nd Netw ork …
8. Phishing attack uses invisible characters to bypass email …
9. Trending Topics – February 20, 2025 – Hunter Strategy
10. Threat intelligence | Microsoft Security Blog
11. The AI Wire | AI Security Incident News Feed
12. Microsoft 365 Device Code Phishing Campaign Bypasses Password Theft With Legitimate Login Flow
15. Evolving Threat: Microsoft AiTM Phishing Attacks