Phishing gang hides invisible characters inside emails to slip past filters
Microsoft says attackers are splitting bait words like 'funding' with unprintable Unicode tags so security scanners never see them.
Key points
- Microsoft's Security Research team has warned of a high-volume phishing campaign using invisible Unicode tag characters to slip past email filters.
- Attackers hide the characters inside financial bait words such as "funding" so scanning tools fail to recognise them as suspicious.
- The trick abuses Unicode Tags, a block of characters (U+E0000 to U+E007F) that most software renders as nothing at all.
- Human readers still see a normal-looking email, so the lure works even though the underlying text is scrambled.
- Defenders should strip or flag Unicode tag characters at the email gateway and keep training staff to double-check finance-themed messages.
Microsoft has flagged a large phishing campaign that hides invisible characters inside emails to fool the software meant to catch them.
Phishing is when criminals send fake messages that look like they come from a bank, an employer or a supplier, hoping the reader clicks a link or hands over a password. The new twist is in how the emails are built. Words that'd normally trigger an alarm at the mail server, things like "funding" or "payment", have been chopped up with hidden characters. A human still reads "funding" on screen. The filter sees gibberish and waves it through.
What are these invisible characters?
They're called Unicode tag characters. Unicode is the global character standard that lets computers display every language and symbol. Tucked inside it is a small block of codes, U+E0000 to U+E007F, that render as nothing visible. They were designed for language tagging and were later mostly abandoned.
Attackers have rediscovered them. Drop one between the "f" and "u" of "funding" and the word looks fine to a person but reads as "f[nothing]unding" to a scanner checking for finance-related lures.
How is this campaign different?
Invisible-character tricks have appeared before, usually aimed at confusing AI assistants that read hidden instructions inside a document. This campaign flips the idea. The hidden characters aren't there to talk to an AI. They're there to break up words so ordinary email filters can't spot the scam.
Microsoft describes the volume as high, meaning millions of messages rather than a targeted handful. The lures lean on money themes: fake funding notices, invoices, payment confirmations. Classic bait in a new disguise. We covered the related problem of AI-written lures slipping past gateways on 20 August, and the trend since then has been consistent: attackers keep finding ways to wrong-foot filters before defenders can retune them.
Should ordinary people be worried?
Mostly this matters to the people who run email systems, but caution helps everyone. Any unexpected email about money, especially one pushing you to click a link or open an attachment, deserves a slow second look. If it claims to come from your bank or a supplier, contact them through a number you already trust, not the one in the email.
The emails will look normal. That's the whole point. Hover over links before clicking, check the sender address carefully, and treat urgency as a red flag rather than a reason to hurry.
What defenders should do
| Action | Why it helps |
|---|---|
| Strip Unicode tag characters (U+E0000-U+E007F) at the mail gateway | Removes the disguise before the filter sees the text |
| Flag messages containing any character from that block | Legitimate mail almost never uses them |
| Re-run keyword and brand-impersonation rules after normalising text | Catches the lures the attackers were hiding |
| Refresh staff training on finance-themed lures | Humans are the last line when filters miss |
Microsoft hasn't named a specific group behind the campaign. The company is telling defenders to update filtering rules to treat the tag block as suspicious by default. Given that Unicode tag characters have essentially no legitimate place in commercial email, that's a low-cost rule with a high-confidence signal.
