Automating Agent Hijacking Via Structural Template Injection

Agent hijacking, highlighted by OWASP as a critical threat to the Large Language Model LLM ecosystem, enables adversaries to manipulate execution by injecting malicious instructions into retrieved content. Most existing attacks rely on manually crafted, semantics-driven prompt manipulation, which...