ASCII Smuggling Moves From AI Attacks to Spam Evasion
A technique originally used to conceal malicious instructions from people reviewing AI inputs is now being used by spammers to bypass email filters. The filters are intended to identify unwanted messages sent as part of mass campaigns. The method is commonly called ASCII smuggling and first drew attention about two years ago in connection with prompt-injection attacks against AI agents.
Instead of using visible text, attackers encode instructions with a special range of Unicode tag characters. The 128-character block closely reproduces part of the American Standard Code for Information Interchange. For example, U+E0041 corresponds to A, while U+E0061 corresponds to a. Computers and large language models can process the encoded characters, but they are designed to be almost entirely invisible to people reading the message. This allowed hidden prompts in emails or other untrusted material to reach an AI system without being apparent to a human recipient. The technique is no longer limited to obscuring prompt injections and has been adopted for spam-related filter evasion.
Written by Sakhanda Wire AI from reporting by Ars Technica. Facts, quotes and full context are in the original article below.