Verifying the Stack for Research: A Working Session from Campus Cluster to Hyperscal

The debate on advanced AI compute has hung on a false choice: move fast to enable research, or lock systems down to meet export-control and security obligations. This workshop starts from the proposition that research centers must do both, simultaneously and by design. Verification-by-design, building trust, compliance, safety and performance into the stack from the first deployment decision, is the mechanism that makes this possible, whether the environment is a campus cluster or a hyperscale cloud tenancy. This is not theoretical. Live deployments, including KAUST's Shaheen III GPU partition, are running frontier research under this model today. Participants work in mixed tables of researchers, HPC operators, cloud and infrastructure providers, security teams, and compliance staff. Each table takes one research enablement challenge, from moving data into a controlled environment to emerging problems the field has not yet solved, and builds a point of view across people, technology, and process that holds true for both on-premises and cloud deployments. Tables defend their approach against a peer challenge and share it with the room. The session runs under the Chatham House Rule (no attribution to any speaker or organization, and no recording), and anonymized learnings will feed community work on verification standards.