Cybersecurity firm Barracuda has found that a decades-old email trick called "text salting" is successfully fooling some AI-powered spam filters, allowing large volumes of phishing messages to reach inboxes undetected. The technique, long used to defeat traditional secure email gateways, works by hiding harmless-looking filler words within a malicious email in ways invisible to human readers but visible to scanning software, confusing machine-learning and LLM-based tools into misclassifying the message as safe.

Barracuda said it had detected more than one million retail-themed phishing attacks using text salting since April. Attackers typically hide the filler text using one of three methods: CSS cropping, which shrinks the visible window; text manipulation, which pushes the extra content off-screen; or zero-font techniques, which insert invisible words into visible phishing copy. While modern email security systems have generally adapted by stripping out hidden text and flagging suspicious insertions, Barracuda says many AI tools process text plainly without distinguishing hidden from visible content, and it recommends a layered defence covering sender reputation, authentication checks and HTML analysis rather than relying on AI filters alone.