Privacy-preserving collaboration on shared datasets, so research institutions can study together without pooling records or exposing participant data.
Multi-institution research creates value by combining datasets — but participant data, proprietary methodologies, and institutional IP cannot be pooled in a shared cloud environment. Researchers need a way to collaborate without exposure.
The collaboration layer for multi-institution research: run shared AI workloads across institutional boundaries inside TEEs, with each institution's data sealed from the others.
When individual institutions need private AI inference on their own data, Cube AI provides the full platform.
Explore Cube AIPrivacy-preserving research AI enables scientists to train models, run analyses, and share findings across institutions without pooling raw participant data. Instead of transferring records to a shared platform, each institution's data is processed inside a Trusted Execution Environment — hardware that prevents other parties and the platform operator from reading the raw inputs. Only the agreed result is released, with cryptographic proof of what computation produced it.
TEE-based collaboration avoids many cross-border data transfer requirements because no raw data crosses any border — only the computation result does. Participant records stay at each institution; the joint workload runs inside an attested enclave and produces outputs (a model, a statistical result, a risk score) that are shared. This approach can substantially simplify IRB ethics review and GDPR cross-border transfer compliance under Article 44.
Remote attestation is a cryptographic mechanism that produces a hardware-signed proof of exactly which code executed inside a TEE, on which sealed data, with which configuration. For research, this means that any collaborator — or a journal reviewer — can independently verify that the published result was produced by the stated algorithm on the stated data, without the institutions needing to share the underlying records. It is a technical reproducibility guarantee that peer review alone cannot provide.
Yes. Each hospital contributes its patient data to a joint training run inside a TEE. The model sees the combined data; no hospital ever sees another's records. The resulting model can be validated against each institution's data independently. This enables multi-site clinical AI studies that would otherwise require complex data-sharing agreements and data transfer to a central processing site.
TEE-based research AI requires servers with AMD SEV-SNP, Intel TDX, or NVIDIA H100 Confidential Computing support. For most research workloads, this means a small number of TEE-capable nodes for sensitive computation, rather than replacing an entire HPC cluster. Prism AI and Cocos AI provide a free tier suitable for research evaluation — full production infrastructure can be deployed on institutional servers.
Institutional Review Boards typically require that participant data not be shared outside the institution without explicit consent. TEE-based collaboration provides a technical mechanism to satisfy this requirement: raw data is never transferred to another institution. The attestation report, which can be made available to the IRB, proves that the computation respected the data boundary. This is a stronger assurance than contractual protections, which depend on the other parties' compliance.
Talk to the team about multi-institution AI collaboration, privacy-preserving research methods, and free tier access.
