Microsoft says email spammers are adopting ASCII smuggling (arstechnica.com)

Microsoft has reported a significant rise in email spam tactics utilizing a method known as ASCII smuggling, a technique previously recognized for enabling stealthy AI attacks through prompt injections. This approach allows malicious prompts to be embedded within emails using a range of Unicode tags that are processed by large language models (LLMs) but remain nearly invisible to human readers. For instance, a tag like U+E0041 corresponds to the letter “A,” allowing spammers to bypass filters designed to detect unwanted messages. The adoption of ASCII smuggling is noteworthy for the AI and machine learning community as it represents a growing sophistication in how spam is disseminated. Microsoft's Defender for Office observed a staggering increase in detections, from approximately 21,000 to over 2.5 million within a few days this February, highlighting the urgency for advanced filtering systems to combat this evolving threat. The challenge lies in the dual nature of the technique: while it aids in concealing malicious instructions from human oversight, it simultaneously complicates the detection of spam by traditional models, necessitating innovative strategies to enhance email security and protect against such obfuscation methods.