Red teaming, in which specialists adversarially probe: Which periodic assessment activity does this describe?

Adversarially probing a system to provoke harmful or policy-violating outputs is the definition of red teaming.

The question

Preparing for AIGP? Take the free 5-min readiness quiz →

1. Threat modeling, in which analysts map likely attackers, assets and attack paths before any testing.

   Plausible but wrong: threat modeling is a planning activity that identifies risks conceptually rather than actively provoking harmful outputs from the live system.
2. Red teaming, in which specialists adversarially probe the system to surface harmful or unsafe outputs. ✓

   Correct: red teaming is the structured adversarial exercise of deliberately trying to make a system produce harmful, biased or unsafe results.
3. Penetration testing, in which testers exploit infrastructure weaknesses to breach the hosting network layer.

   Plausible but wrong: penetration testing targets infrastructure and network vulnerabilities, not the model's tendency to generate harmful content.
4. Bias auditing, in which reviewers measure disparate outcomes across protected demographic subgroups.

   Plausible but wrong: bias auditing statistically evaluates fairness of outcomes rather than adversarially eliciting harmful or policy-violating responses.

How to remember it

Adversarially probing a system to provoke harmful or policy-violating outputs is the definition of red teaming.

More Understanding How to Govern AI Deployment and Use questions

• To detect performance decline and trigger maintenance: What is the primary purpose of this continuous →
• A post-market monitoring plan describing how the deployed: Which documentation artifact is specifically →
• Define permitted uses and contractually restrict them: Which governance action most directly reduces the risk →
• All 424 Understanding How to Govern AI Deployment and Use questions →