Adversarial Evaluation

Run adversarial evaluation that goes beyond happy-path testing. Learn the threat model (who is attacking, with what capability, with what goal), attack libraries (TextAttack, IBM Adversarial Robustness Toolbox, NVIDIA Garak, Microsoft PyRIT, Anthropic AI Safety Levels evals), defence verification, and the relationship between adversarial robustness and operational reliability under unusual but non-adversarial conditions.