ASCII smuggling crosses over from AI prompt injection to phishing evasion (www.microsoft.com)

Microsoft researchers have uncovered a sophisticated phishing campaign leveraging a technique called "ASCII smuggling," originally popularized in AI prompt injection studies. Attackers utilized invisible Unicode tag characters to break apart words like "funding" in email content, circumventing detection by traditional email filtering systems. This innovative use of a previously academic tactic highlights how adversaries are adapting AI-era strategies for real-world phishing, significantly increasing Microsoft Defender for Office 365's alerts to over 1.3 million messages from just 21,000 the day prior. The significance of this finding for the AI/ML community lies in its implications for cybersecurity. While ASCII smuggling was recognized for its potential to manipulate AI models, its application in phishing emphasizes vulnerabilities in AI-driven spam detection systems. The Unicode Tags block, utilized here, has largely been neglected in prior evasion strategies, presenting a new challenge for developers of email security technologies. This evolving landscape necessitates that defenders recalibrate their filtering mechanisms to ensure they account for these invisible characters, blending human-like understanding with machine-learning efficiency to mitigate phishing risks.