- The Mess: Attackers are using invisible Unicode characters to hide phishing keywords inside text that looks completely normal to the human eye.
Microsoft tracked a high-volume campaign using a technique called ASCII smuggling. Invisible Unicode tag characters were inserted into financial lure words, splitting them for security filters while leaving them visually unchanged for the victim.
A person sees “funding.”
A filter may see something else.
The technique is particularly interesting because ASCII smuggling became widely known as an AI prompt-injection trick.
Now attackers are using the same invisible-text concept against ordinary email security.
- The Damage: Microsoft observed the campaign reaching more than 2.37 million messages in a single day at its peak.
The goal was not to exploit an email client or deliver a new zero-day.
It was simpler.
Break the words security systems are looking for.
If a detection rule expects one continuous keyword, inserting an invisible character can turn that keyword into fragments while the recipient still sees a perfectly normal sentence.
Microsoft says more than 99% of the observed messages were still caught by other layers, including reputation systems, URL analysis, authentication checks and machine-learning protections.
But that’s not the point attackers just proved.
The text your filter sees may not be the text your user sees.
- The Fix: Email security systems should normalize and inspect Unicode before applying keyword, signature and content-based detection.
Security teams should also hunt for unexpected characters from the Unicode Tags block in suspicious messages and avoid relying on literal keyword matching as a primary phishing defense.
Bugstoday’s Opinion
AI security research has a habit of escaping the AI lab.
First, researchers used invisible characters to hide instructions from humans while exposing them to machines.
Now phishers are using the same idea to hide words from machines while showing them to humans.
That’s the interesting part.
No zero-day.
No malware exploit.
No sophisticated cryptography.
Just a character you can’t see sitting inside a word you can.
And somewhere between what the computer reads and what the human sees, the attacker found another gap.
Today’s Bugs. Tomorrow’s Breaches.
Technical Sources
- Microsoft Security Blog — ASCII smuggling crosses over from AI prompt injection to phishing evasion
- Microsoft Defender telemetry and threat research
- Unicode Tags block documentation
- BleepingComputer
