1M+ Emails Use Hidden Text to Dupe AI Security Filters
Hackers are using a simple, age-old technique – text salting – to bypass modern email security filters, including those powered by AI. Researchers at Barracuda Networks observed over 1 million retail-themed phishing emails employing this method, leveraging LLMs to generate and implement the salting techniques more efficiently than ever before. Traditional email security gateways struggle to detect these emails because they focus on visual presentation rather than the underlying content and context.
Artificial intelligence and LLMs are proving surprisingly ineffective against text salting, allowing phishing emails to slip through email security filters. Researchers at Barracuda Networks have observed more than 1 million retail-themed phishing emails since April that utilize hidden text to deceive AI-powered security checks. This tactic, a technique once used to evade older spam filters, is now more effective than ever thanks to the capabilities of large language models (LLMs).
Secure email gateways (SEGs) typically work by checking for straightforward signals, such as whether the sender's domain is legitimate and whether standard authentication protocols are used. However, the emails observed by Barracuda are mixed, with some using standard DomainKeys Identified Mail (DKIM) for authentication, while others utilize compromised but legitimate websites as sender domains. The real challenge for SEGs lies in interpreting the content – looking for malicious or spammy language by filtering for specific phrases, salesman-like language, excessive references to money, shady formatting, and overuse of symbols.
Hackers are circumventing these filters by manipulating the HTML code underpinning the email. They achieve this by peppering the email with innocuous content alongside spam, such as injecting suspicious words like "urgent" or "rewards" with nice words like "puppy" or "book." They can also insert random stories or long passages of unremarkable text to break up spammy paragraphs, and hide text by shrinking it to a font size of zero or containing it inside a viewing window with a height and width of zero.
Modern security tools are attempting to un-hide this hidden text by looking for the HTML or CSS rules that make text invisible to users. However, attackers are layering multiple salting techniques on top of one another, so that if one fails, another might succeed. LLMs are also making these techniques easier and quicker to produce, allowing attackers to generate text and modify code at speeds they couldn't otherwise attain. "Defending against this requires more than scanning for suspicious keywords or obvious indicators," says Gutierrez. "Security tools need to evaluate the full context of the message, including the relationship between the visible content, hidden or excessive text, links, sender behavior, and the action the email is trying to prompt."