1. Which AI is explicitly excluded from the EU AI Act?
✓ Correct — Correct.
Military and national security AI is explicitly out of scope.
2. Google's ATEAC (2019) is instructive because:
✓ Correct — Correct.
Google's ATEAC is a failure case — dissolved in two weeks, illustrating the risks of governance structures without clear authority or stakeholder consultation.
3. An organization that commits to pre-deployment governance but deprioritizes post-deployment monitoring has:
Correct.
Governance that ends at deployment is not governance — it is setup. Real governance is the ongoing work of monitoring, updating, and maintaining control.
4. Mandatory AI incident reporting is modeled on:
✓ Correct — Correct.
Aviation safety's confidential near-miss reporting created a knowledge-building culture that dramatically reduced accidents — AI incident reporting advocates propose an analogous system.
5. Behavioral testing of a vendor AI system can determine all of the following EXCEPT:
Correct. Without model weights, you cannot know feature importance. You can only observe behavior.
Behavioral testing reveals what the system does, not why — you cannot determine feature importance without accessing the model internals.
6. Government social scoring is classified as:
✓ Correct — Correct.
Government social scoring is in the prohibited tier — banned entirely.
7. Consequence mechanisms in AI accountability are important because:
✓ Correct — Correct.
Consequence mechanisms — reputational, financial, or career consequences for governance failures — create the incentives that make accountability commitments meaningful.
8. What does cross-company policy comparison uniquely reveal that single-company analysis cannot?
✓ Correct — Correct. Industry-wide gaps are invisible in single-company analysis — they only appear when you compare across companies and find the same omission everywhere.
Collective gaps are the unique insight from comparison — if every company in an industry has the same gap, it signals a structural blind spot rather than individual failure.
9. Pre-deployment review in AI compliance should:
✓ Correct — Correct.
Pre-deployment review should scale with risk classification — applying identical review to all AI wastes resources on low-risk systems and underprotects against high-risk ones.
10. The strongest positive indicator of substantive AI governance — as opposed to ethics washing — is:
✓ Correct — Correct. Demonstrated blocking authority — actual cases where governance stopped or required modification of a deployment — is the hardest thing to fake and the strongest evidence that governance has real teeth, not just advisory status.
Demonstrated blocking authority is the strongest indicator. Principles documents, team size, and public reporting are all consistent with performative governance. Actual rejections or required modifications are not.
11. The most productive functional definition of AI governance is:
✓ Correct — Correct.
The functional definition — whatever actually shapes decisions — is most productive because it identifies all governance levers, not just the obvious ones.
12. In the IBM Watson for Oncology case discussed in the module, the fundamental governance failure was:
Correct.
The failure was governance-level: the organization did not understand the system or verify vendor claims before deployment.
13. The core distinction between a policy document and a governance system is that a governance system:
✓ Correct — Correct.
The distinction is operational specificity — actors, actions, authority, consequences — not legal status, subject matter, or who develops it.
14. "Process theater" as a form of ethics washing involves:
✓ Correct — Correct. Process theater is governance activity without governance authority — committees that meet, assessments that are completed, but no mechanism to actually stop a deployment based on governance findings.
Process theater is governance without authority — the process exists, the meetings happen, but the governance body cannot actually block or require modification of deployments.
15. The capture problem in internal AI ethics governance is:
✓ Correct — Correct.
Capture means internal governance faces structural incentives to serve organizational interests — because reviewers are paid by the organization being governed.
16. The passive voice test identifies accountability gaps by:
✓ Correct — Correct. "Audits will be conducted" in active voice becomes "Team X will conduct audits and report to Y" — which immediately reveals whether that assignment was ever made.
The passive voice test works by active rewriting — "systems will be monitored" → "who monitors, what metrics, reported to whom?" When no answer exists, accountability was never assigned.
17. Starting policy drafting by identifying failure modes rather than principles tends to produce:
✓ Correct — Correct. Failure-first drafting produces requirements calibrated to actual governance risks rather than abstract values — the result is governance that addresses real failure modes.
Failure-first drafting grounds requirements in the actual harms governance is supposed to prevent, producing more targeted and effective requirements than starting from abstract principles.
18. The NIST AI RMF is non-prescriptive because:
✓ Correct — Correct.
Non-prescriptive design allows the framework to work meaningfully across very different contexts and risk profiles.
19. Both providers and deployers have obligations under the Act because:
✓ Correct — Correct.
Both have separate obligations. A deployer cannot shield itself by claiming it only purchased, not built, the system.
20. Why is binding international AI governance difficult?
✓ Correct — Correct.
Structural barriers prevent binding global AI governance: disagreements on goals, pace of change, and the private company problem.