AI Audits and Third-Party Assurance Quiz

Answers and explanations appear as you take the quiz.

1. Which describes an adversarial external audit?
   • A company's own compliance team reviewing its policies
   • A notified body certifying a product
   • Researchers testing a system through its public interface without the vendor's cooperation
   • A regulator reading a company's internal memo
2. What does a governance or process audit mainly examine?
   • Policies, documentation, roles and risk management
   • Only the model's accuracy on a test set
   • The system's electricity usage
   • User interface design
3. What does NYC Local Law 144 require for automated employment decision tools?
   • A ban on all AI hiring tools
   • Independent bias audits
   • Open-sourcing of model code
   • Government approval of each hire
4. How is the impact ratio used in NYC bias audits calculated?
   • Total hires divided by total applicants
   • False positives divided by false negatives
   • A category's selection rate divided by the rate of the most selected category
   • Model accuracy divided by training time
5. Under the EU AI Act, how do most standalone high-risk systems undergo conformity assessment?
   • Through an internal control procedure by the provider
   • Through a mandatory court hearing
   • Through public voting
   • They are exempt from assessment
6. Which access level gives an auditor training data, weights and code?
   • Black-box
   • Grey-box
   • Sandbox
   • White-box
7. What is audit-washing?
   • Deleting audit logs after review
   • Presenting a narrow audit as a general clean bill of health
   • Running the same audit twice
   • Auditing only public-sector systems
8. Why should auditors record the model version and audit date?
   • Because the law requires monthly audits
   • Because retraining can invalidate the results
   • Because audit fees depend on dates
   • Because older models are always fairer