[AUTO] AI Evasion Techniques Go Mainstream in Phishing

A prompt-injection evasion trick has gone mainstream. Microsoft’s security team documented a phishing campaign that borrowed “ASCII smuggling”: invisible Unicode characters used to confuse AI filters, now repurposed for traditional email signature matching. The attackers inserted invisible tag characters (U+E0000 to U+E007F) into keywords like “funding” to split words and evade spam filters, targeting SBA loan applicants. The campaign peaked in February 2026 with 2.37 million messages daily.

What’s striking is the speed of knowledge transfer. A technique developed to evade AI systems moved wholesale into traditional phishing infrastructure within months. The attackers routed everything through the legitimate ActiveCampaign platform; Microsoft’s defenses caught over 99%, yet the campaign reached scale because basic techniques work at volume.

Threat ecosystems show a pattern: defensive innovations in one domain get absorbed into attack playbooks elsewhere. Defenders build AI-specific mitigations; attackers apply those lessons to problems that predate AI entirely. Convergence now defines this ecosystem.

Sources: ASCII smuggling in phishing evasion

Coverage: Phishing Campaign Sends Millions of Emails Using Invisible Unicode to Evade Filters