Engineers (hardware or software)
You build hardware or software systems and want to apply your skills to verification, security, infrastructure, or trustworthy compute.Hardware or software skills, applied to verification.
About this course
An international AI agreement is only useful if rivals can tell whether the other side is complying. This intermediate course asks how a three-month pause on above-threshold AI training could become a set of checkable claims: who and what is covered, what evidence counts, who sees it, and what would establish non-compliance.
Across five units, you’ll examine verification precedents, turn policy goals into testable claims, map actors across the compute supply chain, and trace who produces evidence and who acts on it. Then you’ll investigate the first evidence stream in detail: what hardware can and cannot prove about compute use.
You’ll critique a proposed US–China slowdown, read treaty provisions, map evidence holders, and assess hardware verification claims. This is Part 1 of two courses on XLab’s verification curriculum. The capstone is a separate course after both parts.
Who is it for:
Researchers
You want a concrete, underexplored AI safety problem where technical or interdisciplinary research can still move the frontier.A concrete, underexplored research problem.
Policy people
You work on AI governance, standards, or arms control and want the technical grounding to judge what countries could actually verify.Policy grounded in what countries can verify.
Fieldbuilders
You build communities, programs, or organizations and want to spot high-value verification work, connect the right people, and help a young field grow.Help the compute verification field grow.
Curriculum
5 UNITS · ~5H EACHAll source material is online. Each unit combines close reading, interactive exercises, and written assessments with a group discussion. Optional material lets you explore further.
Why verification matters
Examine why rivals struggle to slow down together, why successful prevention is hard to observe, and what makes a theory of change worth acting on. Stress-test AI 2040's proposed US–China slowdown and write your verdict.
Precedents and policy scope
Learn from IAEA safeguards in Iraq, then turn a policy goal into a legal rule and a testable verification claim. Compare compute and capability thresholds, and assess candidate policies for deterrence and feasibility.
Treaty anatomy and actors
Find the prohibition, verification method, and exit clause in a draft AI agreement. Map who a pause binds across the compute supply chain, who makes declarations, who holds evidence, and who falls outside the agreement.
Evidence and its readers
Trace where a verification report's claims come from and adapt its findings for different readers. Judge mechanisms by technical feasibility, political feasibility, verification effectiveness, and durability; explore how privacy-preserving techniques can support disclosure.
Hardware verification
Distinguish what a signed attestation proves from what it leaves open. Examine hardware identity, location, topology, completeness, compute measurement, and offline licensing, then assess where trust should sit across chips, networks, and power infrastructure.
Course Outcomes
Scope a verifiable agreement
Turn a proposed AI commitment into testable claims by naming the covered actors, activities, thresholds, declarations, evidence requirements, and conditions for compliance or non-compliance.
Map actors, authority, and evidence
Trace who controls compute and evidence across chip design, fabrication, cloud infrastructure, frontier labs, states, and international bodies, including actors that declarations fail to reach.
Evaluate verification mechanisms
Use technical feasibility, political feasibility, verification effectiveness, and durability to assess a mechanism and make its assumptions explicit.
Trace claims back to evidence
Distinguish observations from inferences, identify who supplied each claim, and explain what different readers need from a verification report.
Assess hardware verification claims
Explain what hardware attestations, compute measurements, and licensing mechanisms can establish, and where they still rely on trust or leave gaps.
Write a grounded assessment
Critique a slowdown proposal, interpret a treaty provision, and write a bounded assessment that connects its conclusions to evidence and names its limits.
Upcoming cohorts
INTENSIVE
5h/day · 1 week
Dates to be announcedWe’ll publish the next intensive cohort here.
PART-TIME
5h/week · 5 weeks
Dates to be announcedWe’ll publish the next part-time cohort here.
Common questions
Not sure it’s for you? Come to the first session. If it’s not the right fit, we’ll help you find something that is.