Deep-XPIA – Prompt injection benchmark for multi-agent AI systems

Deep-XPIA is a security benchmark designed to evaluate the resilience of multi-agent AI systems against prompt injection attacks, where malicious inputs can hijack agent behavior. It is intended for AI security researchers and developers building complex, interconnected agent architectures. The project is interesting because it addresses a critical, emerging vulnerability in collaborative AI systems, moving beyond single-model attacks to test how adversarial prompts can cascade through agent-to-agent communication.