Model Disclosure Requirements

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-safety2-model-disclosure-creators

1. Before approximately 2023, what was the standard practice for AI labs regarding disclosure about their models?

   1. Labs were required to submit detailed technical reports to government agencies
   2. All model weights had to be released under open-source licenses
   3. Labs had to publish their training datasets publicly
   4. Disclosure was entirely voluntary with no regulatory requirements

2. According to current EU AI Act requirements, which of the following must general-purpose AI model providers publish?

   1. Real-time usage data from all deployed models
   2. Names and salaries of all research engineers
   3. Detailed documentation and copyright-policy summaries
   4. The complete source code of their models

3. What additional obligations does the EU AI Act impose on providers of models designated as systemic-risk models?

   1. They must release all model weights for free download
   2. They must open their offices to daily inspections by EU officials
   3. They are prohibited from deploying models larger than a specific parameter count
   4. They face additional obligations including evaluations and incident reporting

4. What did California SB 53 (signed in 2025) require from developers of frontier AI models?

   1. Free access to their APIs for academic researchers
   2. Transparency reports on their safety frameworks
   3. Public release of all training data
   4. Immediate shutdown of models that fail initial testing

5. What type of information did the White House voluntary commitments from 2023 ask major AI labs to provide?

   1. Employee background checks and hiring practices
   2. Complete lists of all GPU hardware used in training
   3. Red-teaming disclosures and content-provenance commitments
   4. Customer data usage policies

6. What are model cards and system cards designed to communicate about an AI model?

   1. The mathematical formulas used in the model's architecture
   2. Intended use, evaluations, limitations, and safety testing
   3. Pricing information for commercial API access
   4. The exact hardware specifications used during training

7. Why does the lesson argue that transparency alone is insufficient for AI safety?

   1. Because most people lack the technical skills to read model cards
   2. Because disclosure regimes need audit mechanisms to verify claims independently
   3. Because model cards contain mostly irrelevant information
   4. Because AI models change too quickly to document accurately

8. The lesson quotes an observation that disclosure might become what if it becomes mandatory but superficial?

   1. Too expensive for small companies
   2. Checkbox-shaped and performative
   3. A competitive advantage for startups
   4. Redundant and unnecessary

9. Which entity developed the concept of Model Cards for Model Reporting in 2018?

   1. OpenAI
   2. Google
   3. Meta
   4. Anthropic

10. What is the relationship between transparency and accountability in AI, as described in the lesson?

    1. Transparency is the first step toward accountability, but both are still under construction
    2. Transparency automatically creates accountability without any other mechanisms
    3. Accountability is no longer relevant with modern AI systems
    4. Transparency and accountability are essentially the same thing

11. What type of information does the lesson say is 'summarized at best, rarely reproducible' in current model disclosures?

    1. Red-team findings
    2. API pricing structures
    3. Model architecture diagrams
    4. Hardware requirements

12. What is the current status of the UK AI Safety Institute's evaluation access requirements?

    1. They were eliminated after Brexit
    2. They are legally binding under UK law
    3. They apply only to UK-based AI companies
    4. They are voluntary but expected from major labs

13. Which of the following is identified in the lesson as a subject of active litigation related to AI development?

    1. Employee hiring practices
    2. Model naming conventions
    3. Copyright-relevant data sources
    4. API rate limiting policies

14. What quality variation does the lesson note about model cards and system cards across different releases?

    1. Quality varies widely between labs and across releases
    2. They are only published for open-source models
    3. They are uniformly required by law in all jurisdictions
    4. They have been discontinued since 2020

15. What category of information does the lesson say is necessary but not sufficient for AI safety?

    1. Social media engagement
    2. Transparency
    3. Academic citations
    4. GPU computation