Mandate Engineering — Foundation

What the programme teaches

Mandate Engineering is a proposed professional discipline concerned with the design, issuance, operation, verification, modification, suspension, revocation and closure of authority granted to synthetic systems. A mandate is a documented and revocable authorisation, issued by an identified principal to an identified actor, to pursue a specified outcome using defined capabilities and resources, within explicit boundaries, evidence obligations, escalation conditions and accountability arrangements.

A mandate is not necessarily a legal instrument. The programme teaches operational authority design; it does not validate a legal power the principal does not possess, replace legal advice, or create legal personhood for a synthetic actor. The term "mandate engineering" is in independent use elsewhere; the Institute does not claim to have coined it or to hold exclusive rights in it.

A successful participant is intended to be able to convert an organisational objective into a documented, bounded, testable, observable and revocable mandate; operate that mandate in a controlled setting; produce sufficient evidence of what occurred; and defend the resulting authority and accountability design before an independent reviewer.

Objects of the discipline

Principal

On whose legitimate authority the mandate rests — not merely whoever asked

Accountable owner

The named human who answers for the mandate while it runs and holds pause and revocation rights

Synthetic actor

The identified system, configuration and identity that receives the authority

Scope

The outcome authorised, the non-goals and the reserved human decisions

Permissions and prohibitions

What is required and permitted; what may not be done, stated so that each prohibition is testable

Delegation

How authority propagates to sub-agents, tools and identities, and where attribution would break

Evidence

What must exist to show what happened — including negative evidence that a prohibited thing did not occur

Limits

Numeric limits with a stop behaviour: budget, calls, time, territory, data

Suspension, revocation, recovery

Mechanism, authority, effect and verification of pausing, ending and recovering a mandate

Governance review

Independent challenge of accountability, residual risk and whether the mandate should continue, change or close

Learning outcomes

1. Diagnose mandate need.
2. Specify a mandate.
3. Design delegation and identity controls.
4. Make the mandate verifiable.
5. Operate a mandate.
6. Escalate and revoke.
7. Govern and review.
8. Defend demonstrated competence.

Who it is designed for

• AI and automation product managers; operations and process designers
• Founders and leaders introducing agentic workflows
• Risk, compliance, audit, assurance and model-risk professionals
• Enterprise and solution architects; technology consultants
• Professionals expected to approve, supervise, investigate or insure autonomous work

Foundation means the first professional level in the Institute's proposed pathway, not elementary AI literacy. No programming qualification is required.

Assessment, in brief

Ten assessed domains scored 0–4; four critical gates (boundaries, evidence, revocation, accountability) that no overall average can offset; a defined list of automatic non-pass conditions; a first assessor, a blind second reading of the gate dimensions, moderation of disagreements, and an oral defence under a live change. All designed at development-draft level; none yet run.

Four fictional teaching cases

C-01 Procurement Support Agent · C-02 Research Digest · C-03 Bounded Financial Authority (simulated funds only) · C-04 Agent Fleet and Nested Oversight. Every organisation, person, document and log in the cases is fictional; laboratories are card-, document-, log- and extract-based simulations with no live agent environment, payments or transactions. Case contents are issued only to participants.

What it is not

• Not an AI-literacy introduction, a vendor-specific agent-builder course or a coding bootcamp
• Not a legal, compliance, cybersecurity, audit, insurance or model-risk qualification
• Not a licence to deploy or approve high-risk systems
• Not an academic degree, accredited diploma or university qualification — No accreditation exists or is claimed; the current architecture is non-degree.
• Does not require or permit uncontrolled real-money authority at Foundation level

When Autonomous Agents Improve Themselves

A proposed teaching case built on C-04 Agent Fleet and Nested Oversight. A fleet of agents that improves its own coordination proposes three changes — a workflow optimisation, a change to its reporting contract and expanded execution authority. Participants decide which the existing mandate covers, who must authorise the rest, what evidence must survive and how the system is paused or rolled back.

It applies the programme’s core questions of scope, evidence and revocation to change that originates inside the system, drawing on a recent preprint on recursive self-improvement and a research note that analyses it.

Who Watches the Agent?

A second proposed teaching case in the setting of C-04 Agent Fleet and Nested Oversight. An engineering fleet’s monitor turns out to have four weaknesses — a coverage gap, a switch within the agents’ reach, a fail-open error path and pressure to resume after an automatic pause. Participants decide what must change, who decides, which evidence is needed and when the system may run again.

It applies the programme’s questions of evidence, oversight and revocation to the monitoring of agents, drawing on METR’s published red-team of an AI developer’s internal agent monitoring.