Tag: AI security

Invisible Unicode tag characters create a new security risk for AI-powered applications by hiding malicious instructions inside seemingly harmless prompts. Because some large language models still interpret these invisible characters as valid text, attackers can use them to manipulate responses or inject harmful commands without user awareness. These characters occupy a unique Unicode block and produce predictable UTF-8 byte patterns, making them detectable at the network layer. By adding a custom WAF rule, the Progress Kemp LoadMaster can identify and block requests containing invisible tag characters, helping protect AI applications from prompt injection and other emerging threats.