Understanding and defending against vulnerabilities like data poisoning, model inversion, and adversarial attacks that target AI systems