AI · private AI
Processed where
nobody can read it
Sensitive inference runs inside isolated, attestable environments, so raw financial data is never exposed to the host running the model — provably, rather than on our word.
01
Isolated
The sensitive paths execute in confidential compute. The operator of the machine cannot read the data being processed on it, and neither can we.
02
Attestable
An institution can be shown cryptographic evidence that its data was processed without a human or third party ever able to see it. “Trust us” is not a control.
03
Reserved, not universal
Confidential compute costs latency, money and model choice. We spend it on the genuinely sensitive paths rather than taxing every feature to say the word in a deck.
Where it applies
ScopeSensitive inference paths only
IsolationAttestable enclave, host cannot read
ProofAttestation available to the client
CostLatency and model choice, accepted knowingly
Everything elseStandard path, still minimised