Shango MID

Shango MID decides whether an automated action is allowed before it happens, and keeps proof of that decision that someone outside the company can check.

Software agents increasingly write to systems of record — creating, changing and deleting things that affect real people. Most oversight of that activity is written afterwards, by the same system that did the thing. Shango MID puts the decision in front of the action, and puts the evidence outside the system's own control.

When an agent tries to write, the request is checked against declared authority and scope before it is allowed. The decision and its evidence are written together in a single database transaction, so a decision that cannot be evidenced does not happen at all. The design also provides for each receipt to be counter-signed by a separate legal entity, so the record can be checked without trusting Shango or the organisation using it. That signing path was exercised across 1,000 governed decisions on 4 June 2026; it is not running in the current build, where receipts are committed but not counter-signed.

It is built for organisations that will one day be asked to account for a specific automated decision nobody flagged at the time — what was allowed, on whose authority, and who is answerable. Regulated sectors first, because that question reaches them soonest.

It does not establish that an allowed action actually took place, or that nothing happened outside the governed path, and counter-signing is not currently running. Those limits are stated separately and remain open.

Shango MID is stewarded by Ishaan Ghosh.

The registered work addresses a growing governance problem created by software agents capable of creating, modifying, and deleting records with real organizational or human consequences. Its focus is not merely retrospective logging. It is designed to place an authorization boundary in front of the proposed write and preserve evidence of the decision.

Shango MID is building a pre-action authorization and decision-evidence layer for automated writes.

When an agent proposes a write, the architecture checks the request against declared authority and scope before allowing it to proceed. The decision and corresponding evidence are designed to be committed together in a single database transaction so that an authorization decision that cannot be evidenced does not simply proceed without a corresponding record.

The architecture also provides for receipts to be counter-signed by a separate legal entity so the evidence can be inspected without relying exclusively on Shango MID or the organization operating it.

Shango MID declares a governance boundary around automated write authorization, declared authority and scope, decision evidence, and independently inspectable receipts.

The registered record states that an external counter-signing pathway was exercised across 1,000 governed decisions on June 4, 2026. It also expressly states that this counter-signing pathway is not operating in the current build.

The registration references an OWASP AISVS C9 scenario-suite result publicly deposited through Zenodo, reporting 12 of 14 cases matching, with the mechanism split identified by case.

Shango MID also states that it was designed with reference to EU AI Act Article 14 concerning human oversight. It does not claim an EU AI Act conformity assessment and does not claim an ISO/IEC 42001, NIST AI RMF, or other framework assessment where one has not been performed.

What Shango MID brings to the table is a serious pre-action governance boundary.

Rather than beginning after an agent has already changed a system of record, Shango MID asks whether the proposed write is authorized before allowing it to occur and attempts to preserve the authorization decision and its evidence together.

That creates a strong point of contact with TA-14 because consequence-bearing execution depends not only on whether an action can technically occur, but on whether authority, evidence, scope, continuity, and the right to proceed remain governable before consequence attaches.

What TA-14 is looking for next is closure across the execution boundary.

The registered declaration already identifies that boundary clearly. An authorization receipt does not itself establish that the authorized action actually executed, that the resulting state corresponds to the authorized action, or that no consequential write escaped the governed path.

A deeper TA-14 review would therefore look for evidence connecting:

authorization → evidence commitment → permitted write → actual execution → resulting state → outcome record.

It would also examine refusal behavior, bypass resistance, authority changes, replay, failed or contradictory evidence, and the current operational status of independent counter-signing.

Those are evidence questions, not adverse conclusions.

Registration preserves Shango MID's attributable declaration, identity, stewardship, version, evidence relationships, claims, limitations, and chronology within the TA-14 AI Governance Registry.

Registration does not certify Shango MID, endorse its effectiveness, establish regulatory conformity, validate ownership or priority, prove technical performance, or establish that every consequential action passes through its governed path.

Its declared strengths and declared limitations remain part of the same permanent record.

Shango MID presents several useful pathways for governed continuation inside the TA-14 AI Governance Exchange.

A future evidence-backed demonstration could test a specific proposed agent write from request through authorization, refusal or permission, receipt creation, actual downstream execution, resulting state, and replay.

TA-14 could also examine the previously exercised external counter-signing pathway and distinguish the demonstrated June 2026 capability from the presently deployed version.

Additional work could include an independent TA-14 review, participant response from the Shango MID steward, technical comments, replication or demonstration requests, evidence challenges, and later evidence submissions.

Any later work should remain separate from the original Registry declaration so the chronology shows what was registered first and what was demonstrated, reviewed, corrected, or strengthened later.

Shango MID illustrates an important distinction in AI governance:

Proving why an automated action was allowed is not automatically the same as proving what subsequently happened.

That distinction is valuable.

Shango MID brings a pre-action authorization and decision-evidence architecture to the Exchange. TA-14 brings the broader question of whether the evidence, authority, continuity, binding, execution, and resulting consequence remain connected across the complete governed route.

The architectures do not need to become the same architecture for that relationship to be useful.

The strongest next step is evidence that shows exactly where the Shango MID boundary ends, where downstream execution begins, and how the two can remain attributable without either layer claiming more than it proves.