Research and explainers on control infrastructure for trusted AI in finance — pooled pricing without disclosure, governed model-risk agents, controlled and attested execution, and the Evidence Pack a reviewer can retrieve.
Pool dealer and investor data into a consensus price for illiquid bonds. Each contributor sees only the result and its own distance from it, and every published price carries its own evidence.
Read the use case Use caseHow an approved, deterministic pricing engine reconciles desk marks against independent prices — tolerance checks, governed sign-off, a valuation reserve, and an Evidence Pack from the same run.
Read the use case PerspectiveSR 26-2 covers the model under test, not the agent drafting the file. E-23 expects alternative controls for autonomous models. Effective challenge has to be enforced outside the model.
Read the perspective ResearchTwo peer-reviewed IEEE papers co-authored by our CTO Chen Feng — atomic exchange across independent systems and heterogeneous-TEE assurance — so a result does not rest on one operator or one vendor enclave.
Read the research note ExplainerWhat data, model, policy, approvals, agent requests, runner receipt, and results go into the record Risk, Compliance, or Audit can retrieve when the run finishes — and what it does not prove.
Read the explainerTell us the workflow you're evaluating and we'll share the relevant material — or set up a technical walkthrough.