§A1 About this framework
Draft (AI)Status & version, how to cite, review cadence, AI-use statement, how it was made (method), credit
DRAFT — NOT FROZEN
This page becomes the citable v1.0 snapshot when every section freezes on 31 August 2026. Until then it renders the living state, watermarked.
Asset Management Council · Asset Zero
The Human–AI Boundary Framework for Asset Management
Working draft towards Version 1.0 (31 August 2026)
Built from 2,796 recorded responses across four webinars (June–July 2026), validated by volunteers across three review cycles.
Co-developers — the Working Group
Named on the framework's cover page and in the snapshot's front matter
Birendra Grewal · Fernan Abuid · Hema Wadhwa · Ivan Beirne · Kai Dong · Lalinda Karunaratne · Martin Boettcher · Oludolapo Olanrewaju · Rick Minato · Russell Bunn · Sardar Khan · Steven McCann · Susan Rebano-Edwards · Tim Davies · Titus Naidoo
Reviewers
Named on the contributors page
Birendra Grewal · Fernan Abuid · Kai Dong · Martin Boettcher · Sardar Khan · Tim Davies · Titus Naidoo
Community contributor: Collective acknowledgment (responses were anonymous and stay that way)
Live roster and how to join: /credit
The 18 sections, as closed
§A2 Why a boundary framework
In review (cycle 1)The one finding ("AI amplifies what you already have; the boundary is a governance choice"); the is/ought gap; who this is for
§A3 The evidence base
Draft (AI)The series (4 webinars, 2,796 responses), method, Lifecycle AI Map figure, reading rules, caveats
§A4 Where this sits
Draft (AI)ISO 55000/55013 · PMI AI-in-project-work standard (Jun 2026, cited) · VAISS/AI6 · NIST AI RMF · EASA tiers · SR 11-7 "effective challenge" · AU anchors: SOCI, FAR, Privacy Act automated-decision transparency (Dec 2026), NSW WHS Amendment (Digital Work Systems) Act 2026 (enacted, awaiting proclamation) · EU AI Act (Annex III high-risk obligations deferred to 2 Dec 2027; Art 14 oversight unchanged)
§B1 The ladder & the human roles
In review (cycle 1)AI role levels 0–5 (0 None · 1 Assistant · 2 Analyst · 3 Recommender · 4 Controlled actor · 5 Autonomous) = the series ladder (no role → assist → analyse → recommend → act w/ approval → act autonomously); the retained-rights set (approve · validate · own value judgement · accept & record risk · stop · override · interpret context · govern knowledge · define purpose · accountable by name); which rights rooms under-price cold
§B2 The seven principles
In review (cycle 1)Each: statement · evidence · live rating · practice notes · status. Draft set = [W5B] §3, restructured so the series' rated principles are visibly inside them (P2⊃W1 3.86 · P3⊃W2 3.76 · P4⊃P3-A/B/C · P5⊃P4-A/B · plus the ledger principle elevated — see §3 below)
§B3 The Decision Rights Matrix
In review (cycle 2)12 AM decision rows × (assist/analyse/recommend/act-bounded) × humans-retain × named-accountability column × max level; populated from boundary-vote data; "never" rows written as concrete work classes
§B4 The boundary test
In review (cycle 2)The nine questions ([W5B] §5), reordered to lead with "what decision is being changed?" and "should this be automated at all?"; scoring/usage guidance; screening tool keyed to asset criticality
§B5 Envelopes, triggers and the stop
In review (cycle 2)Envelope anatomy (scope · magnitude · velocity/rate · reversibility window · conditional no-gos · owner · expiry · edge behaviour); the trigger gate; the engineered stop + drill; the volume budget with the community's draft number (median 7.5 WOs/planner/wk; 30% never) and the EEMUA-191 alarm-budget analogy; the reversibility premium (38pp)
§C1 Records & ledgers
Draft (AI)AI decision record · action ledger (actions AND denials) · deferral ledger · AI-use register · re-signing schedules; ISO/IEC 42001 alignment; why records: the community's highest-rated mechanism
§C2 Failure modes & lessons library
Draft (AI)13 consolidated modes ([W5B] §6) + field prevalence (75% have seen bad-data→confident-output) + case library links; "have you seen this?" instrument for reuse
§C3 Verifiable oversight & capability
Draft (AI)Meaningful vs rubber-stamp oversight; effective challenge (SR 11-7); automation-bias countermeasures; evaluate the human–AI team (GMLP P7); deskilling guardrails
§C4 Vendor & black-box AI
Draft (AI)Procurement clause set (audit rights, warranties, indemnities — vendor-liability stats flagged as law-firm analysis citing Stanford); deployer-not-developer posture; assurance market caveats
§C5 Getting started → maturing
Draft (AI)0–5 maturity path; the is/ought gap as the starting map (23% auto-create today · 36% hour-pause · 22% drilled · weights showable 39% · audit guess 0%); small-organisation / local-government path
§D1 Open Questions Register
Draft (AI)[W5B] §9 ten questions, merged with per-webinar leaving-questions; each: status · source attribution · what would settle it (OQ2 = the number; OQ5 = reversibility premium interpretation)
§D2 Recommendations to members
Draft (AI)The nine ([W5B] §8), prioritised by the community in the cycle-3 vote (W5 confirms/showcases)
§D3 Glossary
Draft (AI)Ladder verbs, envelope, engineered stop, rough consensus, provenance labels (verified/inferred/unknown), etc.
§D4 Credits & provenance appendix
Draft (AI)Contributor tiers; comment-disposition archive; Asset Zero boundary telemetry (from Governor's ledger); AI-use statement (long form)
D4 — Provenance appendix
Evidence base: 663 + 524 + 625+28 + 956 = 2,796 recorded responses · four webinars, June–July 2026.
| Ask | Cycle | Volunteers answered |
|---|---|---|
| Confirm your part in August (RSVP + hours) | 0 | 15 |
| MA1 — Judge each principle, and pick the structure | 1 | 11 |
| MA2 — Break one principle | 1 | 10 |
| MA3 — The missing principle | 1 | 10 |
| Reviewer packet 1 — "The Confident Wrong Answer" | 1 | 6 |
| B3 — Decision Rights Matrix review | 2 | 11 |
| B4 — Apply the boundary test | 2 | 11 |
| B5 — Envelopes, trigger and stop | 2 | 11 |
| Final pass — the whole-document last call | 3 | 0 |
| Final pass — rank the Recommendations for Members | 3 | 0 |
0 disposition rowsadopted — every volunteer input answered in writing, signed per row. The full logs render on each closed section and export per cycle. The project’s own AI ran under pre-registered boundary levels; its action ledger is public at /progress.
AI synthesises the community's 2,796 responses, drafts sections with line-by-line provenance, and compiles the disposition logs. Humans set direction, judge content, run the calls, sign the sections and own the outcome. The project itself runs under the framework it is producing: the AI's permissions are declared in advance, every action and refusal is logged, and the log is published. A full AI-use statement appears in the framework's front matter.
© Asset Management Council — community-developed guidance carrying the community’s authority, not a formal standard. Print this page for the citable PDF.