A method for concealing malicious content known as ASCII smuggling has shifted from being primarily used in attacks on artificial intelligence systems to becoming a widespread spam evasion technique, according to recent findings from Microsoft.
The technique relies on a block of 128 Unicode characters that are invisible to human readers but detectable by computer systems. These special characters can replace standard letters while appearing identical to ordinary text on screen. The mechanism gained initial prominence around two years ago as a way to embed hidden instructions into AI model prompts without detection by human reviewers.
Beginning in early February, Microsoft Defender for Office recorded a dramatic increase in spam messages employing this approach. Daily detection signatures jumped from approximately 21,000 to over 1.3 million within a single day, eventually climbing to 2.5 million within four days. This surge continued for several months before subsiding sharply in mid-May.
Spammers are using these invisible characters to obscure keywords commonly flagged by email filters, such as specific dollar amounts and terms like “credit” and “term.” By inserting Unicode tags within words like “funding,” the text appears normal to recipients while filters may register fragmented segments instead. This approach targets both literal string-matching systems and more advanced machine learning and natural language processing-based filters that modern email security relies upon.
While spammers have historically used other invisible characters such as zero-width spaces to bypass detection, the adoption of Unicode tags likely reflects their novelty to existing filter systems. Microsoft indicated that the primary advantage lies in circumventing sophisticated AI-driven classification models, which may fail to recognize the intended words when tokenized or processed by standard email analyzers. The company provided recommendations for developers to strengthen filter programming against this emerging technique.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI