Text Salting: Why Hidden

Text salting is a technique where large blocks of harmless text are hidden within an email to evade AI models by presenting a benign surface appearance while the full content remains visible to the system. This evasion works because automated systems read the entire raw content, whereas human recipients only see the rendered portion of the email. The attack leverages this gap: a short, suspicious lure is presented in visible text, while the majority of the message consists of filler material that dilutes any immediate threat signal for classifiers.
The effectiveness of this technique has grown significantly, with the use of text salting paired with link cloaking increasing nine-fold over six months in confirmed phishing cases. Furthermore, attackers employ specific "hard hides"—CSS techniques like `display:none` and zero font size—to ensure the hidden text is invisible across different rendering modes. A crucial element of this evasion involves a second trick where visible links direct automated scanners to decoy domains, creating two separate, benign signals that confuse external analysis.

* Text salting hides a large block of harmless text inside a phishing email.
* The technique is used so an AI model reads a mostly benign document while the human sees only the lure.
* Hard hides used include `display:none`, `opacity:0`, and zero font size positioning text off-screen.
* This evasion works because an AI classifier reads all raw content, while a human reads the rendered email.
* The hidden content is often a large block of prose, such as a fake newsletter or chat thread.
* In 84% of cases, the hidden text traveled with a second trick: a link pointing to a decoy domain that cloaks the actual destination for automated scanners.
* Attackers consistently use hard hides rather than soft hiding techniques.
* The technique is associated with toolchains emitting both content dilution and link cloaking.
* The incidence of this pattern grew nine-fold between December 2025 and June 2026 among confirmed evasive phishing cases.

The narrative shifts from simple content deception to industrial-scale adversarial behavior, demonstrating that evasion is now an engineering discipline rather than an isolated trick. The core pattern is the consistency of structure: bulky, non-link text hidden using hard CSS properties, deployed alongside a separate link cloaking mechanism. This structural regularity indicates a toolchain is at work, suggesting these are not the actions of individual operators but the output of automated phishing kits.
The implication for defense is that focusing solely on content analysis fails because attackers have learned to manipulate the input space for models. The shift requires moving defenses upstream to inspect the structure of the email—distinguishing between what renders and what is concealed. This necessitates a layered defense that incorporates dedicated detectors focused not just on semantic content but on the physical presentation and relationship between visible and hidden elements.
The growth rate observed in these techniques points toward an evolution where adversary tactics are optimized for automated systems, forcing a paradigm shift from reactive signature updating to proactive structural anomaly detection. The focus must move from detecting specific text patterns to identifying persistent adversarial structures that enable bulk content manipulation across evolving AI evaluation methods.
Bridge Questions: If detection focuses on the structure of what is hidden rather than just the visible content, how can security systems evolve to grade the intent of both visible and concealed information simultaneously? What structural indicators exist beyond CSS properties that reliably signal malicious obfuscation in general email rendering pipelines?