Microsoft reported a massive increase in spam messages using ASCII smuggling, with the number of detected signatures jumping from 21,000 per day to over 1.3 million in early February. The technique, which hides malicious prompts in Unicode tags, has now become a common evasion method for spammers.
ASCII smuggling works by encoding malicious instructions in a range of Unicode tags that are invisible to humans but readable by computers. These tags mimic parts of the American Standard Code for Information Interchange, making them difficult to detect by standard filters.
Spammers are embedding Unicode characters to evade filters that search for keywords like 'credit' and 'term' in mass emails. By inserting invisible text into words like 'funding,' they can trick filters into misreading the content while leaving the visible text intact.
"Because tag characters are invisible to humans but exist at the text-processing level, the same property that makes them useful for smuggling instructions into a model also makes them useful for obfuscating keywords before a detector evaluates them," Microsoft explained. The company emphasized that the mechanism is similar to prompt injections, but the intent is inverted.
The surge in ASCII smuggling highlights the growing challenge for spam filters as machine learning models become more prevalent in email classification. Microsoft warned that standard classifiers may fail to detect such attacks unless they use OCR or account for Unicode normalization.
Microsoft did not specify how spammers are acquiring the Unicode tags, and the open question remains about how to effectively counteract this evolving threat. The company is now providing guidance on how developers can improve filters to detect ASCII smuggling.
Source: arstechnica
