Adversarial testing (red-teaming) is best described as?

Adversarial testing (red-teaming) is best described as?

• A. Adversarial testing (red-teaming)
• B. Hallucinations
• C. Ethical Bias
• D. Exact Match Metrics

Answer: (A)

Adversarial testing, or red-teaming, is the practice of deliberately probing an AI system with crafted attempts to break it, bypass safeguards, or reveal safety and security weaknesses. The goal is to expose how the model behaves under adversarial pressure—identifying prompt injections, data leakage risks, or policy violations—so that governance, monitoring, and protective controls can be strengthened before real-world deployment. This is distinct from issues like hallucinations, which describe the model generating plausible but false information; from ethical bias, which concerns biased behavior or outcomes; or from exact-match metrics, which measure task accuracy rather than resilience to adversarial manipulation. By conducting adversarial testing, teams gain practical insight into where defenses fail and how to improve risk controls.

Adversarial testing, or red-teaming, is the practice of deliberately probing an AI system with crafted attempts to break it, bypass safeguards, or reveal safety and security weaknesses. The goal is to expose how the model behaves under adversarial pressure—identifying prompt injections, data leakage risks, or policy violations—so that governance, monitoring, and protective controls can be strengthened before real-world deployment. This is distinct from issues like hallucinations, which describe the model generating plausible but false information; from ethical bias, which concerns biased behavior or outcomes; or from exact-match metrics, which measure task accuracy rather than resilience to adversarial manipulation. By conducting adversarial testing, teams gain practical insight into where defenses fail and how to improve risk controls.