Center for AI Safety Compute Cluster (web)
Citation
Center for AI Safety (CAIS). “CAIS Compute Cluster.” Program page, date not stated. https://safe.ai/work/compute-cluster. Read 2026-11-05.
Bottom line
- CAIS gives “free compute for research projects in ML safety” from “a cluster of 80 A100 GPUs, with a dedicated team to provide support to cluster users”.
- The page says “We are currently not accepting new applications for external access to the cluster.” It lists 16 collaborating professors, among them David Wagner and Dawn Song at UC Berkeley and Cihang Xie at UC Santa Cruz. It lists about 60 papers made with the cluster.
Background
The page describes the cluster, who may use it, collaborators and the research produced. It is a small, free, domain-restricted compute program run by a nonprofit, with a use rule (ML safety) rather than an institutional rule.
Key claims
- “Conducting useful AI safety research often requires working with cutting-edge models, but running large-scale models is expensive and often cumbersome to implement. As a result, many non-industry researchers are unable to pursue advanced AI safety research.” (Overview)
- “To address this issue, CAIS runs an initiative to provide free compute for research projects in ML safety, based on a cluster of 80 A100 GPUs, with a dedicated team to provide support to cluster users.” (Overview)
- “~100 AI safety research papers produced. 150+ AI safety researchers actively using the cluster. 2,500+ Research citations” (Overview)
- “We are currently not accepting new applications for external access to the cluster.” (Who is Eligible for Access?)
- Collaborators include “David Wagner, Professor of Computer Science, University of California Berkeley”, “Dawn Song, Professor of Computer Science, University of California Berkeley” and “Cihang Xie, Assistant Professor of Computer Science, UC Santa Cruz”. (Our Collaborators)
Evidence and method
The program’s own page. Figures are self-reported and undated. The paper list marks each item “Under review for conference”. That label is stale: the page itself says DecodingTrust won an award at NeurIPS 2023.
Limits
No cost, funding source, allocation rule or application form is shown. The page does not say when applications closed or if they will reopen.
⚠️ Contradiction: Center for AI Safety About (web) says “>100 researchers” have used the compute since March 2023; this page says “150+ AI safety researchers actively using the cluster”. Both are undated self-reports.