Cybercriminals are using invisible Unicode characters to hide phishing lures inside finance-themed emails, a technique that has allowed large-scale campaigns to slip past some security filters, according to research from Microsoft. The tactic, known as ASCII smuggling, has previously appeared in AI prompt injection attacks, where malicious instructions are encoded with Unicode characters from the Tags block, a range of code points that are visually invisible. Microsoft threat researchers identified a specific campaign that relied on this obfuscation method to disguise words such as "funding," "capital," and "loan" by inserting an invisible character in the middle of the term, breaking the string so that keyword-based detection systems fail to recognize it.
The campaign peaked in late February, with Microsoft detecting up to 2.37 million malicious messages per day. Although the daily volume dropped sharply after May 15, the operation continued, with the high-volume phase lasting roughly three months from February 9. Microsoft noted that these dates reflect only the period when the technique was observed in its telemetry, not the full duration of the underlying campaign, which began earlier without the Unicode trick and persisted after its use declined. The company stated that the method worked as intended, allowing the messages to evade filters that rely on word lists, though its Defender for Office 365 still blocked more than 99 percent of the emails based on other signals, including sender reputation, IP addresses, and domain checks.
The phishing messages were sent through infrastructure tied to ActiveCampaign, a legitimate email-marketing platform. Microsoft flagged a cluster of 148 finance-themed sender domains on February 9, which accounted for approximately 96 percent of all messages that its hunting logic marked for Unicode tag signatures. These domains used terms like "funding," "capital," "loan," "advance," and "credit," and the emails promoted business funding, loan, and credit services. After Microsoft reported the abuse, ActiveCampaign stated that its moderation systems treat invisible Unicode characters the same way they handle regular text, flagging heavy use of such characters as suspicious.
For defenders, Microsoft recommends stripping or normalizing Unicode tag characters and other invisible code points before applying keyword, regex, or signature-based detection. The company also advises treating unexpected tag-block characters as a strong anomaly indicator. Applying the same normalization process to email content before it is passed to AI assistants could reduce the risk of prompt injection attacks, a growing concern in enterprise environments where large language models process incoming messages.
The findings tie into broader measurement of security defenses, as Microsoft鈥檚 Blue Report 2026 evaluates techniques across 338 million simulations run in customer production environments. The report highlights that overall prevention scores can mask what happens after initial access, noting that when attackers use valid credentials, the effectiveness of prevention measures drops sharply. This context underscores that while the Unicode obfuscation trick is notable for evading content filters, it is just one layer of a campaign that already leverages other methods, including legitimate infrastructure and credential-based access.
More cybersecurity news from TechManNews.
