We're building an attested compute layer for healthcare's hardest collaboration problems — where neither party can afford to share their data, and contracts alone aren't enough.
A pharma company has a proprietary imaging or genomics model. A hospital has the patient data. Neither will move first. Federated learning helps with training — but inference on real patients leaks both ways. We think attestation closes the gap.
Sponsors need blinded reads. Sites can't share raw scans laterally. Core labs become a bottleneck and a single point of trust. Decentralized blinded reads — with cryptographic proof of what ran where — reshape the workflow.
National programs and disease consortia want to collaborate but can't legally co-locate data. Federation helps. Hardware attestation gives the audit story regulators in EU, Singapore, India, and elsewhere increasingly ask for by name.
Sharing raw records across institutions
What ran. On whose data. With what version.
Memory, gradients, intermediate state at runtime
A reasonable answer in 2018. The threat model has moved on.
Federation alone leaks gradients. Weights can be reconstructed.
With attestation, the host institution sees only ciphertext.
Trust is contractual. Audit logs come after the fact.
Cryptographic proof that exactly the agreed code ran.
"Data didn't move" — defensible, but increasingly thin.
Hardware-rooted evidence aligned with EHDS & FedRAMP.
Most platforms focus on training collaboration only.
Designed for production at hospital sites and trial centers.
Five fields. We'll reply within two business days with a few times. If the fit isn't there, we'll say so plainly.