Hackers Use Invisible Unicode Characters to Bypass Email Security Filters, Microsoft Warns

ASCII is a character encoding standard that converts readable characters into numeric values for computers. The system can also generate characters that are not displayed on screen but are still readable by machines. Attackers exploit this by inserting invisible Unicode characters into keywords commonly flagged by spam filters, such as "funding", "credit", or "loan". The visible word remains intact to the human eye, but the filter sees a broken string, such as "fun[a long string of characters]ding", which prevents the email from being automatically moved to spam.

Microsoft's report says the technique has been adapted from prompt injection attacks, where ASCII smuggling is used to deliver malicious invisible prompts in emails. When a victim asks an artificial intelligence agent to summarize an email, the AI may process hidden instructions that could range from extracting sensitive data to deploying malware.

The campaign has been active for months. It peaked in February 2026, when Microsoft observed more than 2.3 million such emails per day. Since then, activity has declined but remains ongoing. In early February, Microsoft identified a cluster of nearly 150 sender domains themed around finance, which accounted for almost all (96%) of the ASCII smuggling spam flagged by Defender for Office 365.

Defenders can counter the technique by normalizing Unicode tag characters and other invisible code points before applying keyword detection. Microsoft also recommends treating any unexpected tag-block characters as suspicious.