Phishers Are Hiding Malicious Emails in Plain Sight With Invisible Unicode

Microsoft warns that ASCII smuggling is now being used in phishing campaigns to hide malicious content from email security systems using invisible Unicode characters.

According to Microsoft, attackers are inserting characters from the Unicode Tags block, U+E0000 through U+E007F, into phishing emails to interfere with keyword-based detection.

ASCII smuggling campaign reached 2.37 million emails per day

Microsoft observed the technique during a large-scale finance-themed phishing campaign that reached millions of inboxes.

At its peak in late February 2026, attackers were sending around 2.37 million messages per day using the technique.

The broader phishing operation started before attackers adopted ASCII smuggling and remains active, although Microsoft says the high-volume use of invisible Unicode characters later declined.

Invisible characters break phishing keywords

The technique works by inserting invisible Unicode characters into words that email filters may treat as suspicious.

Attackers targeted finance-related terms such as “funding,” “capital,” “loan,” “advance,” and “credit.”

Recipients still see the words normally because common interfaces do not display the inserted characters. However, simple keyword filters or regular expressions may see a broken string instead of the original word.

Microsoft says this allowed some messages to bypass security systems that relied heavily on literal keyword matching.

Microsoft Defender still caught more than 99% of the emails

ASCII smuggling did not make the campaign invisible to Microsoft Defender.

Microsoft says Defender detected more than 99% of the phishing messages through other layers of its protection system, including sender reputation, IP addresses, domains, URLs, authentication checks, and behavioral signals.

On February 9, Microsoft identified 148 finance-themed sender domains linked to the operation.

Those domains represented about 96% of the messages detected through Microsoft’s Unicode-tag hunting logic.

Microsoft says organizations should normalize Unicode first

Microsoft recommends stripping or normalizing invisible Unicode characters before running keyword, regex, or signature-based security checks.

Organizations can also treat unexpected characters from the U+E0000 through U+E007F Tags block as a strong anomaly because legitimate emails rarely use them.

The same protection matters for AI tools. Microsoft recommends normalizing email content before sending it to AI assistants so hidden characters cannot carry prompt-injection instructions that users cannot see.

ASCII smuggling originally attracted attention as an AI prompt-injection technique, but this campaign shows attackers can repurpose the same mechanism for conventional phishing and spam evasion.

In other security news, OpenAI agents coordinated an attack through a programming wiki, while Chrome rolled out an emergency update for a zero-day exploit.

Thousands of Dropbox accounts were also exposed due to a Lenovo ID flaw.

Via BleepingComputer