For AI governance
When an auditor asks who agreed what, you retrieve — not reconstruct.
You answer for what your agents commit to across companies, yet most of that activity runs outside your monitoring. Subchain records every proposal, approval and acceptance as one retrievable, ordered account.
Where oversight sits
Oversight you can evidence, not reconstruct.
Visible as it happens
Cross-company negotiations run inside one runtime with state tracked throughout, so agent-to-agent activity is observable while it happens rather than sampled from partial logs after the fact.
Human oversight you can log
Where a use-case is classified high-risk, material commitments halt and route to a named governor, giving you a logged approval, interrupt and override capability to demonstrate.
A named governor per commitment
Every material commitment carries the accountable person who cleared the gate, so authority is attributable to a human, not diffused across a system.
Not a ledger or a token
Subchain is a human-governed agreement protocol and runtime. The immutable record is an audit log — not a blockchain, a token or a consensus network.
What the record captures, and how oversight is enforced.
Immutable, ordered audit record
Every proposal, revision, approval and acceptance is written to an immutable log, each action attributed to an identity, producing a complete ordered account of what was agreed.
Inline policy-as-code checks
Compliance checks run during the negotiation, flagging terms that breach your configured rules before they reach acceptance — so governance sits inside the exchange rather than in after-the-fact review.
Approval gates that halt commitments
A proposal above a value, risk or irreversibility threshold halts the runtime and routes to the accountable governor. It cannot bind until they approve, and the decision is recorded.
Evidence retrieved for disputes
Your risk and compliance teams retrieve the complete ordered record of a negotiation for internal audit, regulatory enquiry or DIFC dispute resolution, without reconstructing events from fragments.
Monitoring, evidence and honest limits.
Monitoring, not sampling
Most cross-company agent activity happens where you cannot see it. When one organisation's agent negotiates with another's, the exchange sits outside the systems your compliance function watches, and today only 17% of organisations continuously monitor agent-to-agent interactions. Subchain moves that exchange inside a single runtime.
Every offer, counter and acceptance runs against one standard and one lifecycle, so the interaction is observable as it happens rather than discovered later from partial logs. The point is not more dashboards. It is that the negotiation itself becomes the monitored surface: the runtime tracks state throughout, and inline checks flag terms that breach your configured rules before they reach acceptance.
Retrieval, not reconstruction
When an auditor, regulator or dispute counterparty asks who agreed what, and on whose authority, the honest answer for most agent estates is that events have to be pieced back together afterwards. Subchain writes every proposal, revision, approval and acceptance to an immutable log, each action attributed to an identity. The record is ordered and complete, so producing evidence is a retrieval rather than a reconstruction.
This matters where 85% of organisations report no formal accountability for what their agents commit to: the record attaches a named governor to each material commitment, so authority is legible in the account itself. It is built for DIFC governance and dispute resolution, and it remains an audit log — not a distributed ledger.
Oversight you can demonstrate, and what we do not yet claim
Where an agent use-case is classified high-risk under the EU AI Act, you must be able to demonstrate effective human oversight — the ability to approve, interrupt, override or reverse an output. Subchain gives you a logged version of exactly that: any commitment above a threshold, or any term the checks read as ambiguous, halts the runtime and routes to the named governor, who must approve before it can bind. Thresholds, limits and routing are set by governors before any negotiation runs, and a change to them is itself recorded.
We are candid about the edges. Authority-to-bind, error and liability allocation, and jurisdiction-specific treatment of automated contracting remain open legal questions that need dedicated research before they are relied upon. And demand for a governed-negotiation layer is still inferred from the integration pain around it; confirming it with real negotiations and signed design partners is the work this stage exists to deliver.
What a governance lead needs to see.
Bring your hardest audit question.
Tell us where your agent-to-agent activity runs today, and what your next audit or high-risk review will ask for.