MARQUE.
A letter of marque was a commission authorizing a private party to act with delegated sovereign authority — defined scope, defined term, and accountability running back to the grantor. It is the historical instrument for the problem autonomous finance has not yet solved: how to delegate consequential authority to an actor you do not directly control, and remain answerable for what it does.
MARQUE is a reference framework organized around eight questions an autonomous financial agent must be able to answer before capital is exposed to it. It is free to adopt, free to cite, and versioned so that conformance means something specific.
Existing regulation governs models. None of it governs an agent with a mandate.
Model risk management asks whether a model is fit for purpose. Disclosure regimes ask what was told to whom. Conflicts rules ask whose interest was served. All three assume a human decides and a model advises.
An autonomous agent inverts that. It researches, constructs, and executes — and the human contribution moves upstream, into the grant of authority itself. That grant is the governed object, and almost nothing has been written about it.
MARQUE does not attempt to be a control framework, a certification scheme, or a substitute for regulation. It is a question set: the eight things that must have answers, stated precisely enough that the absence of an answer is visible.
An institution that cannot answer all eight for a given agent has not deployed a governed system. It has deployed an ungoverned one with documentation.
Eight questions autonomous finance must answer.
Each question is mapped to the published standards that touch it — and, more usefully, to the ones that do not.
Authority
What is the agent permitted to commit, and who granted that?
Every autonomous action is an exercise of delegated authority. The grant has to be explicit, bounded, revocable, and recorded — or there is no principled answer to what the agent was allowed to do.
Answered operatively byThe Agent Authority Grant
Oversight
Who is accountable in the interval between decision and review?
Human-in-the-loop is not a control if the loop closes after the trade settles. The governed quantity is the gap, not the review.
Drift
How is behavioral change detected before it becomes loss?
Models degrade silently. The failure mode is not a wrong answer, it is a gradually less right one, arriving faster than quarterly review.
Answered operatively byThe Kill-Switch Covenant
Auditability
Can a decision be reconstructed months later, exactly?
Not logged — reconstructed. Same inputs, same model state, same output. Anything less is a narrative about a decision, not the decision.
Accountability
When it goes wrong, who answers?
Liability does not distribute itself across a stack. Someone is the named person, and the framework has to say who before the loss, not after.
Identity
What is the agent, legally, and to whom does its output belong?
An agent that generates a strategy raises a joint-authorship question wearing a fintech costume. Nobody has resolved who owns the output, and chain of title for the model behind it is largely unestablishable today.
Answered operatively byModel Provenance Representations
Systemic behavior
What happens when many agents converge on the same trade?
Correlated automated behavior is a market-structure question, not a firm-level one. Individually compliant agents can be collectively unstable.
Agent-to-AgentIOSCO Supervisory Toolkit for AI Use in Capital MarketsDORA
Answered operatively byThe Stressed Conditions Annex
Standards & interoperability
What does conformance actually mean here?
Without a conformance definition, 'AI governance' is a claim rather than a property. That is what this framework exists to fix.
Answered operatively byThe Schedule Set
What conformance means, and what it does not.
A system is MARQUE-conformant at v1.0 when it can produce, on request and without preparation, a documented answer to each of the eight questions for a named agent in a named deployment.
That is a deliberately low bar and it is met by almost nothing today. It says nothing about whether the answers are good. It is a completeness test, not a quality test — and completeness is the thing currently missing.
- Free to adopt. Cite the version. No permission required, no fee, no registration.
- Not a certification. Nobody is accredited to assess MARQUE conformance, and any claim otherwise is unauthorised.
- Versioned. The question set will change. A conformance claim without a version is meaningless.
MARQUE is an editorial framework published by the QIS Ecosystem. It is not legal advice, it is not a regulatory standard, and it has no official status with any authority.