AI spam filters are getting suckered by old-school text salting

Barracuda reports a widespread phishing technique called "text salting," where attackers insert hidden, benign-seeming words (using CSS cropping, text manipulation, zero-fonts) to confuse ML/LLM-based email filters and evade detection; the company observed over one million retail-themed phishing messages since April and recommends layered email defenses (sender reputation, auth checks, HTML rendering analysis, and visible vs hidden content checks) instead of relying solely on AI filters.