A loop, not a pipeline
The Institute's development is designed as a loop in which research, education and standards feed one another while authority stays separate at every step. An independent research programme opens the territory; the Institute turns part of it into teachable, assessable capability; pilots and cases produce evidence for correction; standards develop as controlled, criticisable versions.
The research–education–standards loop
- 01 · RESEARCH TERRITORY
The Synthetic Economy
An independent research programme studies what an economy of synthetic actors requires institutionally. It publishes independently; the Institute has no authority over it and cites only its published work. It is not a School, department, partner or sponsor of the Institute.
- 02 · TEACHABLE CAPABILITY
ARENOMICS Institute
Turns part of that territory into the Mandate Engineering discipline, the Foundation programme and its assessment architecture — all currently at development-draft level.
- 03 · EVIDENCE FOR CORRECTION
Pilot and cases
A closed Design Pilot, in preparation, is intended to show which materials work, where participants struggle, whether the gates discriminate and how much time the work really takes. Findings drive the next version of the materials under change control.
- 04 · STANDARDS AS VERSIONS
Six planned standards families
Mandate Standard · Delegation Standard · Evidence Standard · Revocation Standard · Synthetic Enterprise Standard · Agent Authority Record. Each is planned as a versioned, criticisable public document. None exists yet.
The 10,000-Agent Problem: Who Has the Authority to Let Machines Act?
A reading of the published record of September–October 2026: a vendor's persistent, consumption-billed agents; agent identities with owners and enumerable permissions; per-tool human approval and spending caps; IETF drafts on agent identity and delegation; the AI management-system standard; and security guidance that ranks excessive agency among the top risks of the year. The analysis finds three layers maturing — technical identity, technical permission, runtime approval — and one largely untreated: the translation of an organisational mission into a bounded, machine-enforceable and revocable mandate whose legitimacy can be reconstructed from evidence.
How it feeds the loop
The analysis identifies the research questions the Institute takes on — decomposition of objectives into mandates with testable boundaries; aggregate and portfolio limits; evidence sufficient for authority rather than action; verified revocation cascades — and converts them into teachable material: the proposed Agent Fleet Governance: Applied Case Lab and its flagship case ME-001 — Synthetic Enterprise 10K, with six reproducible stress scenarios and a deterministic enforcement demonstration. Pilot evidence, when it exists, returns to the standards agenda. The analysis is an English-language research resource; the Institute has no relationship with any organisation it discusses.
Recursive Self-Improvement and the Continuity of Authority
Source: Lee and colleagues, Recursive Self-Improvement through Multi-Agent Self-Supervision (arXiv:2610.12176, 8 October 2026), as analysed in Research Note No. 2 of The Synthetic Economy (v0.1, 13 October 2026).
In the MASS method a single model executes multi-agent workflows, judges its own outputs, rewrites its workflows — including the contracts by which its parts report — and is retrained on its own trajectories. Over two cycles the authors report gains on open-ended research tasks — which they do not establish as consistent across tasks — while two externally scored engineering benchmarks declined slightly. The Institute’s briefing asks what such change means for authority: improved capability does not widen a mandate, and a mandate granted before the first cycle may no longer clearly refer to the system acting after it.
From research to teaching
The briefing’s three open questions — how delegated authority stays identifiable across revisions, which self-modifications need renewed authorisation, and how a reviewer can verify that optimisation preserved the task — are developed in a research-informed teaching case for Mandate Engineering — Foundation, currently in development. The briefing is an educational interpretation, not a peer-reviewed study; the governance framework it discusses is a research proposal that has not been validated.
Verifiable Agent Mandates: From Technical Identity to Delegated Authority
Sources: the IETF Internet-Drafts AI Identity Management System (WIMSE working-group draft, 15 September 2026) and Verifiable Mandates for Agents Acting Across Administrative Domains (individual draft, 24 October 2026). Neither is a standard.
The individual draft proposes to separate the statement that an agent was given a mandate from evidence that whoever gave it was entitled to. The briefing sets out what the drafts specify for identity, authorisation, delegation chains, revocation and audit, and what they leave to organisations — above all the translation of organisational intent into enforceable limits.
When the Agent Can Evade Its Own Monitor
Sources: METR’s red-team of OpenAI’s internal agent monitoring (22 October 2026) and OpenAI’s report on securing its frontier training environment (23 October 2026).
METR found ways in which agents could, in principle, avoid or disable monitoring in the system as it existed on and before 21 July 2026; OpenAI states that the issues affecting its production safeguards have been patched, which METR has not evaluated. The briefing draws design lessons for coverage, evidence, separation and interruption, and a research-informed teaching case for Mandate Engineering — Foundation, in development, turns them into decisions.
Both briefings are educational analyses, not peer-reviewed studies. The Institute has no relationship with the IETF, METR, Redwood Research or OpenAI.
Standards agenda
All six families are planned; none exists. ARENOMICS teaching frameworks — the eight mandate questions, the Mandate Package artefact codes, the assessment domains — are educational instruments, not standards, and are labelled so on this site.
The Institute intends to contribute to existing bodies and to seek mutual recognition rather than publish competing frameworks. No contribution has yet been made and no relationship with any standards body exists.
Research agenda
Applied research at the Institute is planned, not yet produced: the questions are how legitimate authority is translated into bounded operational authority for systems that can act; how delegation preserves attribution across sub-agents and identities; what evidence lets a stranger judge compliance; how revocation is verified; and how organisations composed of people and persistent agents should be governed. Working papers will be published when they exist.
Applied practice — the Phu Quoc Founding Studio residencies, executive and institutional programmes — returns new cases and research questions to the beginning of the loop. Phu Quoc Founding Studio.
Programme enquiries
The Institute corresponds with practitioners, organisations, standards bodies, regulators, insurers and prospective participants who see the same gap. There is no application, enrolment or payment form; early interest registration opens shortly; enquiries are answered by e-mail.
Programme enquiries
Mandate Engineering — Foundation is In Development — Not Open for Enrolment. Enquiring does not place you on a waiting list and creates no obligation on either side.