Cybersecurity firm Barracuda has reported a significant increase in phishing attacks employing "text salting," an older technique now being used to bypass AI-powered email filters. Since April, Barracuda has detected over one million retail-themed phishing attempts utilizing this method. While text salting has historically been effective against traditional secure email gateways, Barracuda indicates it is now also confusing machine learning and large language model (LLM) based security tools.
Text salting involves embedding numerous random, harmless-looking words within a malicious email. The goal is to make the email appear less suspicious to automated scanning systems, leading them to misclassify the message as benign and allow it to reach the recipient.
Attackers typically employ specific methods to conceal this "salted" text from human readers while keeping it visible to automated scanners. These methods include CSS cropping, which restricts the visible window to hide filler text; text manipulation to move the extra content outside the visible screen area; and zero-font techniques, which insert misleading words between legitimate-looking content that is visible to machines but not to humans. The result is an email that appears as intended to a human but is perceived as less malicious or even gibberish by a machine.
While modern email security systems have largely adapted to these techniques, with newer tools capable of identifying and removing hidden text or flagging emails with excessive hidden content, Barracuda states that AI-driven filters have not kept pace. The company's report explains that text salting can "confuse AI-driven content analysis engines by flooding the email with random terms that encourage the AI system into making an incorrect classification decision."
Barracuda further elaborated that LLMs are typically designed to process email text and source code plainly, without inherent understanding of whether text is visible or hidden from a user. Although these models can be trained to recognize such distinctions, it appears most current tools do not incorporate this capability by default.
To counter the influx of these "salted" spam messages, Barracuda advises enterprises to adopt a multi-layered approach to email security. This strategy should move beyond sole reliance on keyword detection and incorporate checks for sender reputation, authentication results, embedded URLs, HTML-rendering techniques, and discrepancies between user-visible and hidden content. The firm also suggested that organizations might consider alternatives to AI-only spam filters.
