Verification

Sign in to track your progress and save your writing.

1. Module 0: Why Are We Teaching This?

   The oldest problem in arms control, applied to AI: when you sign a mutual agreement, how do you know the other party will uphold it? Opens with the welcome and the course's own framing, then the case that ASI risk warrants an agreement at all, why successful prevention is invisible, the intuitions a verification regime runs on, and seven decades of arms-control precedent.

   • 0.0 Welcome
   • 0.1 Introduction: Why Should You Care About AI Verification?
   • 0.1.1 Why Securitizing AI Is Difficult but Necessary
   • 0.1.2 We Need More Theories of Change
   • 0.2 Building Verification Intuitions
   • 0.3 History, Precedents, Parallels
   • Optional: 0.4 Strategic Foundations

   Start moduleOverview
2. Module 1: Policy Scoping & Actors

   What kind of policy are we trying to verify, and who does a treaty rely upon, apply to, and constrain? Compute versus capability thresholds, the effectiveness/feasibility pair, the anatomy of a pause agreement, and the actor map across the compute supply chain.

   • 1.0 Introduction: What Kind of Policy Are We Trying to Verify?
   • 1.0.1 Drawing the Line: Compute vs. Capability
   • 1.0.2 Policies Must Be Effective and Feasible
   • 1.1 Anatomy of a (Pause) Agreement
   • Optional: An International Agreement to Prevent the Premature Creation of Artificial Superintelligence
   • 1.2 Actors: Who Does the Treaty Rely Upon, Apply to, and Constrain?
   • 1.2.1 Geographic Supply-Chain Map
   • 1.2.2 Who Can Prove What
   • 1.3 Upstream and Downstream
   • 1.3.1 Context Distiller

   Start moduleOverview
3. Module 2: Verification Infrastructure and Evidence Streams

   The four buckets of mechanism — hardware, cloud, intelligence, and human evidence — each judged by the claims it can test, the evidence it produces, what it costs to implement, and how it fails.

   • 2.0 What Makes a Verification Mechanism Effective?
   • 2.0.1 Privacy-Preserving Mechanisms
   • 2.1 Hardware
   • 2.1.1 What Evidence Proves Compliance?
   • 2.1.2 From a Chip to a Trusted Statement
   • 2.1.3 Accounting for Hardware: Identity, Location, Topology, and Completeness
   • 2.1.4 Measuring and Classifying Use
   • 2.1.5 Authorization, Licensing, and Control
   • 2.1.6 Where Should Trust Live?
   • Optional: 2.1.7 Reconstructing a Declared Training Run
   • 2.1.8 Policy Judgment: What Role Should Hardware Play?
   • 2.2.1 Provider Records and Workload Observables
   • 2.2.2 Customer Identification and Ongoing Monitoring
   • 2.2.3 Detection Gaps and Policy Limits
   • 2.2.4 Interpreting Cloud Evidence
   • 2.3 Intelligence
   • 2.3.1 Open-Source Intelligence
   • 2.3.2 Imagery and Geospatial Intelligence
   • 2.3.3 Measurement and Signature Intelligence
   • 2.3.4 Financial and Procurement Intelligence
   • 2.3.5 Signals and Cyber Intelligence
   • 2.3.6 From Signal to Intelligence Assessment
   • 2.3.7 Summary and Written Output
   • 2.4 Human Evidence
   • 2.4.1 Insiders and Human Sources
   • 2.4.2 Reporting and Protection
   • 2.4.3 Audits and Inspections
   • 2.4.4 Institutions and Policy Judgment

   Start moduleOverview
4. Module 3: Architecture and Limitations of Low-Trust AI Compute Verification

   How the requirements of an international AI agreement might be translated into a technical verification system: the assumptions, design choices, and unresolved problems involved in verifying AI compute use under conditions of limited trust between the parties.

   • 3.0 A System Overview for Near-Term, Low-Trust AI Compute Verification

   Start moduleOverview
5. Module 4: Capstone: What Would Be Enough for a Three-Month Emergency Pause?

   Putting it all together — then the feasibility judgments the capstone runs on: the four metrics, and how to do the research they ask for, from a practising AI governance researcher. Then the capstone itself: layer the imperfect mechanisms into a regime you can defend, and say where to go from here.

   • 4.0 Putting It All Together
   • 4.1 Feasibility Judgments
   • 4.1.1 How to Do Research Well
   • 4.2 Capstone Project
   • Congratulations

   Start moduleOverview