Two hard problems under high-assurance execution

An AI agent can propose a waterfall, a mark reconciliation, or a screen. When the result has to be relied on outside the institution that produced it, the claim reduces to two stubborn questions:

  • Coordination. When a transaction touches several independent domains — two banks, a desk and its product-control function, an issuer and a trustee — how do you make them act atomically without a single trusted intermediary holding everyone's data and discretion?
  • Assurance. If computation runs inside a Trusted Execution Environment, why should a counterparty (or a regulator) trust the output of one vendor's enclave, given that any single hardware platform can have flaws?

The easy answers — trust the operator, or trust one vendor’s enclave — are exactly the concentration of trust a regulated institution cannot accept. Both papers start from the opposite assumption: the trusted hardware itself may fail.

Mercury — committing across independent systems

Mercury studies atomic exchange across two independent blockchains: either both legs of a trade settle or neither does. That all-or-nothing property is known to be impossible in an asynchronous setting without some trusted party. Mercury replaces a trusted operator with a group of operators running consensus inside TEEs, so clients do not have to stay online, and adds an on-chain challenge-response path so funds remain recoverable if the TEEs become unavailable. In the authors’ evaluation it reduced on-chain costs by roughly 45–68% against XClaim, ZK-bridge, and Tesseract.

The transferable idea for regulated workflows: parties that do not share infrastructure can reach one committed outcome without handing their state to a middleman, and the design keeps a recovery path when the trusted component fails. In OpenBITS that idea sits under the optional shared-state checkpoint — used only when another party must verify the same approved version.

The practical payoff: coordination becomes a protocol with a recovery path, not a trusted middleman.

TeeRollup — assurance from heterogeneous hardware

TeeRollup addresses assurance. Its threat model assumes some TEEs may be compromised or unavailable. Transactions execute on a distributed set of sequencers running heterogeneous TEEs — for example Intel SGX, Intel TDX, AMD SEV, and ARM TrustZone — so the system stays secure even if a certain proportion of them are compromised, and a challenge mechanism keeps funds redeemable if TEEs go offline. The prototype, on the Ethereum Sepolia testnet, reduced on-chain verification cost by about 86% against ZK-rollups, with withdrawals in minutes.

For OpenBITS, the principle carries into the attested tier: a high-assurance result should not rest on one hardware vendor. When execution moves into attested environments, acceptance can require agreement from attesters on different hardware lineages, so one bad enclave cannot push a number through on its own.

From paper to the control plane

Research results don’t become a product by themselves. Most deployments start without enclaves — with workload identity, bank rules, exact approval binding, and signed runner receipts. These papers inform the higher-assurance tiers above that baseline:

  • Deterministic execution. Approved valuation, waterfall, and screening engines produce reproducible output, so anyone with the committed inputs can recompute a result.
  • Stable digests. Inputs, code measurements, and outputs are hashed with a consistent canonicalization, so coordination and attestation operate over comparable commitments.
  • Evidence at execution time. Material steps emit an Evidence Pack — the record a reviewer can check without touching raw data.

The bank keeps its existing systems of record and its authority. What changes is the path a number takes: computed under control, shared with evidence, and — where the threat model calls for it — without resting on one operator or one hardware vendor.

References

  • [1] X. Wen, Q. Feng, J. Niu, Y. Zhang, and C. Feng, “Mercury: Practical Cross-Chain Exchange via Trusted Hardware,” IEEE Transactions on Dependable and Secure Computing, 2025. doi:10.1109/TDSC.2025.3630656
  • [2] X. Wen, Q. Feng, H. Lyu, J. Niu, Y. Zhang, and C. Feng, “TeeRollup: Efficient Rollup Design Using Heterogeneous TEE,” IEEE Transactions on Computers, 2025. doi:10.1109/TC.2025.3596698

Want the technical detail?

The papers are public. We walk institutional reviewers through how they map to the OpenBITS architecture.

Get in touch