1. In the November 2023 OpenAI board crisis, why was Sam Altman reinstated within 96 hours of being fired?
Correct. Microsoft's job offer to Altman and the near-total employee threat to resign demonstrated that financial and employment leverage overwhelmed the board's governance authority — even a structurally unusual safety-oriented governance mechanism.
Microsoft's offer to Altman and nearly the entire staff threatening to follow him created overwhelming financial and employment pressure that forced the board to reverse course and reinstate Altman.
2. Which of the following is a requirement under China's Generative AI Measures (August 2023) with no direct equivalent in the EU AI Act?
3. What is the logic behind "compute governance" as a regulatory approach to frontier AI?
Correct.
Incorrect. Compute governance targets the hardware layer — specialised chips and data centres — because training frontier models requires extraordinary concentrations of specific hardware that can, in principle, be tracked and regulated.
4. What makes the OECD AI Principles' primary practical value as an international governance instrument?
Correct. As non-binding soft law, the OECD Principles' primary function is normative anchoring — giving different national frameworks a common vocabulary, which facilitates mutual recognition and reduces compliance friction.
Not correct. The OECD Principles are non-binding. Their primary value is establishing shared vocabulary and normative reference points that make it easier for different national frameworks to recognise each other's approaches.
5. Under the EU AI Act, which of these domains is listed in Annex III as high-risk?
Correct. Employment decisions are explicitly listed in Annex III as a high-risk domain requiring conformity assessment, technical documentation, and human oversight.
Incorrect. Employment AI is a high-risk Annex III domain. Spam filters and video game AI fall into the minimal-risk category with no specific obligations.
6. US Executive Order 14110 on AI directed the NIST AI Safety Institute to be established within which agency?
Correct. EO 14110 directed the creation of the AI Safety Institute within NIST, which was later renamed the AI Safety and Security Board after the order was rescinded.
Incorrect. EO 14110 established the AI Safety Institute within NIST (National Institute of Standards and Technology), which developed AI safety standards and the AI Risk Management Framework.
7. Canada's Artificial Intelligence and Data Act (AIDA) includes a five-year parliamentary review requirement. This is an example of which adaptive governance mechanism?
Correct.
Incorrect. A mandatory parliamentary review after a specified period is a sunset/review clause — a mechanism that embeds re-evaluation into the legislation itself rather than leaving it to political initiative.
8. The Seoul AI Safety Summit in May 2024 produced what notable commitment from major AI developers?
Correct.
Incorrect. The Seoul summit produced voluntary commitments from major developers — including Anthropic, OpenAI, Google DeepMind, and Meta — to provide safety institutes with pre-deployment model access for evaluation. This remained voluntary, not binding.
9. OpenAI's Superalignment team was co-led by Jan Leike. When Leike resigned publicly in May 2024, his specific critique was:
Correct. Leike's exact resignation statement included: "Safety culture and processes have taken a back seat to shiny products" — an unusually direct public critique of organizational priorities from a departing safety team lead.
Leike's public statement was: "Safety culture and processes have taken a back seat to shiny products." It was remarkable for its specificity and the seniority of the person making it.
10. The "Brussels Effect" in AI governance refers to:
Correct. The Brussels Effect describes how EU regulations de facto globalize because multinational companies find it more efficient to apply EU-level compliance universally than to segment products by jurisdiction.
Incorrect. The Brussels Effect is the documented phenomenon where EU regulations become effective global standards because multinationals apply them universally rather than maintaining jurisdiction-by-jurisdiction compliance architectures.
11. The US National AI Initiative Act of 2020 primarily did which of the following?
Correct. The National AI Initiative Act created coordination structures for federal AI R&D — a research investment and coordination measure, not an AI governance or restriction statute.
Not correct. The National AI Initiative Act of 2020 established coordination mechanisms among federal AI research programs. The NIST AI RMF came later, in January 2023.
12. ISO/IEC 42001 is best described as:
Correct. ISO/IEC 42001 is a management system standard — like ISO 9001 for quality or ISO 27001 for security — that organizations can be certified against. It specifies governance processes, not prohibited applications or technical benchmarks.
ISO/IEC 42001 is a certifiable management system standard, comparable to ISO 9001 (quality) or ISO 27001 (security), specifying organizational processes for AI governance rather than prohibitions or benchmarks.
13. The Frontier Model Forum's $10 million AI Safety Fund was established to:
Correct. The $10 million AI Safety Fund was established to support AI safety research as part of the FMF's stated mission of advancing safety for frontier models.
The AI Safety Fund was established to support AI safety research — part of the FMF's stated mission alongside developing best practices and sharing knowledge with policymakers.
14. Which of the following best describes the core principle of adaptive AI governance?
Correct.
Incorrect. Adaptive governance builds in mechanisms for updating rules as evidence accumulates, rather than fixing them at enactment.
15. The July 2023 White House voluntary AI commitments were signed by how many companies, and which was NOT among them?
Correct. Seven companies signed: Amazon, Anthropic, Google, Inflection, Meta, Microsoft, and OpenAI. Apple was notably absent.
Seven companies signed — Amazon, Anthropic, Google, Inflection, Meta, Microsoft, and OpenAI. Apple did not participate.
16. How does the EU AI Act address the problem of foreign-based AI providers serving EU users?
Correct.
Incorrect. Like GDPR, the EU AI Act applies extraterritorially — any provider placing an AI system on the EU market or affecting EU users must comply, regardless of where the provider is incorporated.
17. Which US statute, predating the AI era, has been cited as the primary legal basis for the FTC's enforcement actions against AI companies making deceptive capability claims?
Correct. Section 5 of the FTC Act — prohibiting unfair or deceptive acts or practices — is the FTC's primary tool for AI enforcement, applied to false capability claims, synthetic endorsements, and misleading AI marketing.
Incorrect. The FTC applies Section 5 of the FTC Act — its longstanding consumer protection authority — to AI-related deceptions without requiring new AI-specific legislation.
18. Colorado's SB 24-205 (2024) is significant in US AI governance because:
Correct. Colorado SB 24-205 became the first comprehensive state AI law in the US in May 2024, covering AI systems used in high-stakes decisions about insurance, employment, housing, and credit.
Incorrect. Colorado's SB 24-205 (May 2024) was the first comprehensive state AI law in the US — it covered AI in consequential decisions and was modeled loosely on the EU Act's approach.
19. What was ProPublica's central finding about COMPAS in its 2016 "Machine Bias" investigation?
Correct.
Incorrect. ProPublica's finding was about differential false-positive rates: Black defendants were mislabelled as likely to reoffend at roughly twice the rate of white defendants.
20. How many AI/ML-enabled medical devices had the FDA cleared in the United States as of 2024?
Correct. The FDA had cleared more than 950 AI/ML-enabled medical devices by 2024, making it by volume the most active US regulator in the AI product space, applying existing 510(k) and De Novo pathways.
Incorrect. As of 2024, the FDA had cleared more than 950 AI/ML-enabled medical devices — by volume, one of the most significant AI regulatory records of any US agency.