New AI threat types: Attacks like prompt injection, data poisoning, and symbolic language exploits target AI systems in ways firewalls can’t detect. Why red-teaming matters: Simulated adversarial ...