How scammers hide the word 'funding' inside emails your filter cannot read

According to Microsoft, a phishing campaign is breaking up financial trigger words like "funding" using invisible characters, so that email filters cannot read them as a whole word. The email still looks completely normal to a person reading it. That is the point.

The technique is called ASCII smuggling. It relies on a block of Unicode characters, U+E0000 to U+E007F, that were built decades ago for language tagging and are now mostly unused. Most fonts and inboxes do not display these characters at all. A person opening the email sees nothing odd. But the characters are still there in the underlying text, sitting inside words, splitting them apart at the code level even while the screen shows a clean, intact sentence.

Where this technique came from

This trick did not start in phishing. It became well known over the past year in a different field: testing how AI assistants can be tricked. Researchers found that hidden instructions could be buried in a webpage or document using these same invisible characters. A human sees nothing. An AI assistant reading the raw text decodes the hidden characters anyway and can be steered into following instructions it was never meant to follow.

Phishing operators have now taken that same property and turned it toward a different target: not an AI model, but the filters that scan email for dangerous words. A word like "funding" is a common flag in scans for financial fraud. Insert invisible characters into the middle of it and the filter no longer recognises the word, even though a human reader's eye fills in the gap without noticing anything missing.

That is the mechanism worth sitting with. The scam does not need to fool the reader's eye at all. It only needs to fool the software checking the email before it reaches the reader. The deception happens at a layer most people never think about, the raw character data behind the words on screen, not the words themselves.

Microsoft's telemetry shows hits on a hunting signal built to catch this technique rose sharply from February 9, 2026, and stayed elevated on weekdays for roughly three months. That pattern, weekday spikes over a sustained stretch, points to an operation run like a business, not a one-off test.

There is a useful detail buried in how these messages were actually caught. Microsoft says most of the flagged messages were stopped by layered protections working together, not by any single Unicode-specific detector. No one filter caught this on its own. It took several checks stacked on top of each other, which is the same principle that should guide anyone reading a suspicious email: do not rely on one signal, whether that is a spam folder, a spelling check, or a gut feeling about tone. A financial request that feels slightly off is worth a second channel of verification, a phone call to a known number, before any money or details move, regardless of whether the email looked clean.