Title page 1

Contents 3

Executive Summary 2

Introduction 4

Report Scope 6

Guiding Questions 6

Selection Criteria 7

Limitations 7

Proposal Case Studies 8

Industry: OpenAI’s Approach to Frontier Risk 8

Civil Society: AI Now Institute's Zero Trust AI Governance 10

Academia: Frontier AI Regulation: Managing Emerging Risks to Public Safety 12

State Government: California's Safe and Secure Innovation for Frontier Artificial Intelligence Models Act 14

Federal Government: Framework to Mitigate AI-Enabled Extreme Risks 18

Analysis 20

Deriving Assumptions 20

Shared Assumptions 21

What Risks Are Important to Mitigate and Who Should Have Primary Oversight of Frontier AI? 22

Who Is Delegated Tasks and Able to Play a Role? 23

Would the Proposed Mechanisms or Tools Actually Achieve the Proposal's Objectives? 25

Unique Assumptions 31

Discussion 33

Conclusion 35

Authors 36

Acknowledgments 36

Appendix A: Assumption Interdependencies 37

Endnotes 38

Tables 6

Table 1. Guiding Questions 6

Table 2. Key Components of OpenAI's Approach to Frontier Risk 9

Table 3. Key Components of AI Now Institute's Zero Trust AI Governance 11

Table 4. Key Components of Academia's Frontier AI Regulation: Managing Emerging Risks to Public Safety 13

Table 5. Key Components of SB-1047 16

Table 6. Key Components of Framework to Mitigate AI-Enabled Extreme Risks 19

Table 7. Shared Assumptions About Which Risks Are Important to Mitigate and Who Should Have Primary Oversight of Frontier AI 22

Table 8. Shared Assumptions About Who is Delegated Tasks and Able to Play a Role 24

Table 9. Shared Assumptions About Which Techniques or Categories of Techniques Are Effective 26

Table 10. Shared Assumptions About Which Processes or Frameworks Are Necessary 28

Table 11. Shared Assumptions About Which Information or Actions Are Useful 30