CrowdStrike identifies five new prompt injection techniques that could compromise AI systems, highlighting the growing risks of adversarial attacks on LLMs.