ASCII smuggling phishing has emerged as a powerful way for cybercriminals to bypass traditional email security filters.
Microsoft security researchers uncovered a large campaign that uses invisible Unicode characters to disguise financial keywords inside phishing emails. The technique allows messages to appear normal to recipients while confusing automated detection systems.
Microsoft found that the campaign reached millions of messages during its peak. The activity shows how techniques originally associated with attacks against artificial intelligence systems are now being adapted for conventional phishing.
How ASCII Smuggling Works
ASCII smuggling uses invisible characters from the Unicode Tags block.
These characters normally do not appear on a user’s screen. However, computers and AI systems can still process them as part of the underlying text.
Attackers take advantage of that difference.
For example, a phishing email might contain a financial word such as “funding.” Criminals can insert invisible Unicode characters into the middle of the word.
The recipient still sees the familiar word. However, a basic security filter may interpret it as a completely different sequence of characters.
As a result, the filter can miss a keyword that would normally trigger a spam or phishing warning.
Hackers Target Financial Keywords
Microsoft found that the campaign mainly used financial-themed messages.
Attackers inserted invisible characters into words associated with business loans, funding and credit offers. These terms are often important signals for email security systems.
The campaign also used large numbers of disposable, finance-themed sender domains.
Microsoft identified roughly 150 domains associated with the main activity. Together, they accounted for about 96% of the messages flagged by the specific ASCII-smuggling detection signature.
Millions of Phishing Emails Sent
The campaign became particularly active in February 2026.
Microsoft’s telemetry showed around 21,000 messages containing the relevant Unicode pattern on February 8. One day later, the figure jumped above 1.3 million.
The campaign then reached more than 2.3 million messages in a single day, with the highest recorded weekday volume reaching about 2.37 million on February 26.
The activity followed a striking pattern. Messages surged during weekdays and dropped sharply over weekends.
Microsoft observed the high-volume phase for roughly three months. It then declined sharply after May 15, although the broader phishing campaign continued to evolve.
ASCII Smuggling Comes From AI Attacks
The technique gained attention through AI security research.
In prompt injection attacks, criminals can hide instructions inside invisible characters. A person may see ordinary text, while an AI system receives additional hidden instructions.
Attackers can potentially use those instructions to manipulate an AI assistant into performing unwanted actions, depending on the system’s capabilities and protections.
The latest campaign uses the same basic technique for a different purpose.
Instead of hiding instructions from people, criminals are hiding parts of phishing keywords from automated email filters.
Traditional Filters Face a New Challenge
Invisible characters are not completely new in phishing.
Cybercriminals have previously used zero-width spaces, soft hyphens, non-breaking spaces and similar techniques to break up suspicious words.
However, Microsoft said the use of the Unicode Tags block at this scale is notable. The characters became better known through AI security research, making their appearance in traditional phishing campaigns an important development.
This creates another challenge for security teams because attackers can continually change the way they disguise malicious content.
Microsoft Recommends Better Detection
Microsoft says organizations can reduce the risk by normalizing text before security systems analyze it.
Security teams should remove or handle invisible Unicode characters before checking messages for suspicious keywords.
Unexpected characters from the Unicode Tags block should also receive additional scrutiny.
Microsoft recommends combining these checks with other security measures instead of relying on one detection method.
AI and Email Security Are Now Connected
The campaign highlights an important change in cybersecurity.
Techniques developed to attack AI systems can quickly move into traditional cybercrime. At the same time, conventional phishing tactics can be adapted to target AI-powered security tools.
That means companies need to protect both the human recipient and the systems that process incoming messages.
Microsoft continues to use multiple layers of protection, including sender reputation, URL analysis, bulk-mail detection, anti-phishing models and content analysis.
ASCII Smuggling Phishing Threat Remains Active
The sharp decline in the campaign’s highest-volume phase does not mean the threat has disappeared.
Microsoft stressed that its telemetry covers the period in which this particular ASCII-smuggling technique was observed. The broader phishing operation existed before the technique appeared and continued after its use declined.
Therefore, organizations should not treat the drop in volume as the end of the threat.
The discovery instead shows how quickly cybercriminals can adapt. As email filters become more sophisticated, attackers are finding new ways to make malicious messages look harmless to both people and machines.
For businesses, normalizing hidden Unicode characters and strengthening phishing detection can help close one of the latest gaps exploited by ASCII smuggling phishing.
