Microsoft researchers identified a widespread phishing campaign that used invisible Unicode characters to split finance-related words and evade email filters. The campaign peaked at 2.37 million messages per day, although Microsoft Defender for Office 365 blocked more than 99% of them based on other signals.
Threat actors used a technique known as ASCII smuggling to hide phishing words inside email messages by inserting invisible Unicode characters between the letters of words that security systems rely on to detect suspicious messages. According to Microsoft researchers, the campaign peaked at approximately 2.37 million messages per day in late February 2026 and remains active, although its volume has gradually declined since May.
How Does the Trick Work?
The attacker inserts an invisible character into a finance-related word, causing a word such as funding to become a form containing a hidden character between its parts. The message may appear normal to the reader, but scanning systems that rely on word lists or text matching may not treat it as the complete original word.
The technique is associated with Unicode characters in the Tags block, within the range U+E0000–U+E007F. The idea has also been used in prompt-injection attacks against artificial intelligence models to hide malicious instructions from users.
Campaign Size and What the Monitoring Revealed
On February 9, Microsoft identified a group of 148 sending domains associated with finance-related themes. They accounted for approximately 96% of the messages detected by the new search capability in Defender for Office 365 for Unicode Tags signatures. The domains used words such as funding, capital, loan, advance, and credit, while the messages promoted business financing, loan, and credit services.
The high-volume phase continued for approximately three months after February 9, then declined sharply after May 15, 2026. Microsoft explains that these dates define the period during which use of the specific technique was observed in its data and do not necessarily represent the beginning or end of the broader campaign.
The report stated that the messages passed through infrastructure associated with the ActiveCampaign email marketing platform. After being notified of service abuse, ActiveCampaign said its moderation systems detect invisible characters in the same way they detect unobfuscated text, and that heavy use of them is treated as a suspicious signal.
Why Does This Matter?
The campaign shows that relying solely on keywords or regular expressions leaves a practical vulnerability to manipulation of text representation. Microsoft recommends normalizing Unicode characters, or removing Tags characters and other invisible characters, before applying keyword, regular-expression, or signature rules.
However, bypassing this check alone was not sufficient to reach users in most cases. Microsoft reported that Defender blocked more than 99% of the messages based on other indicators, including the sender, IP address, domain, and source reputation. The company also recommends treating the presence of unexpected characters from the Tags block as a strong sign of anomalous behavior and applying the same normalization before passing email content to AI-based assistants to reduce the risks of prompt injection.