Ascii smuggling becomes spam tool as AI security flaw exploited

A method once used to attack AI systems, known as ASCII smuggling, is now being utilized by spammers to bypass email filters. The technique allows malicious instructions to be hidden within seemingly harmless text, making it difficult for AI systems to detect while remaining invisible to human readers. The practice, which gained attention two years ago, involves using Unicode tags that mimic standard ASCII characters but are undetectable to humans. These tags, such as U+E0041 for 'A' and U+E0061 for 'a', enable attackers to embed harmful prompts within emails or other content. This method has been adopted by spammers to evade detection mechanisms on email platforms. The technique exploits the ability of large language models to process these hidden instructions, allowing malicious content to slip through security filters undetected. ASCII smuggling works by encoding malicious prompts in a way that is readable by computers but nearly invisible to users. This has led to a growing concern among cybersecurity experts, as the technique continues to evolve and pose new challenges for AI security. The use of such methods highlights the ongoing arms race between attackers and defenders in the realm of AI-driven security systems.