Executive Summary

The short answer

Not on the evidence presently available, under the demanding test this series has adopted. Australia enters 2026 with substantial institutional, scientific, fiscal, operational and community capacity. It is not starting from zero. Emergency services work; governments plan and fund recovery; national climate, demographic, defence and cyber assessments exist; care and social-support systems serve large populations; elections and parliamentary institutions remain established; and AI is already embedded in business, government and critical systems.

Those assets matter. The strongest feasible non-AI reforms—better prevention, reliable public services, adequate workforces, supply redundancy, regulation, conventional digital systems, transparent monitoring, trusted public communication and representative deliberation—could materially improve resilience by 2035. In several functions they offer a credible route to adequacy.

But this report could not demonstrate that the launch-baseline systems plus that strongest feasible reform package would meet every mandatory outcome and continuity test across four interacting risks, seven non-substitutable domains and six operating conditions. Because the method forbids averaging, strong performance in one area cannot cancel a failure in another. Ecological deterioration and democratic or information failure are vetoes, not small deductions. Distributional failures cannot be hidden inside a national mean. Plans, spending and adoption do not count as delivered capability unless evidence connects them to continuity and outcomes.

The result is therefore a provisional comparator gap: P1, the launch baseline plus the strongest feasible non-AI reforms, is not demonstrated adequate as a whole. This is deliberately different from an AI-necessity finding. A gap in P1 identifies functions for Report 2 to test. It does not show that AI can close them, that AI is preferable to further conventional reform, or that its human, ecological and democratic effects are acceptable.

Why the answer is not simply “yes”

Four patterns carry the finding.

First, Australia is stronger at anticipating and responding than at proving prevention, recovery, learning closure and implementation. Emergency management is organised around prevention, preparedness, response and recovery, and the 2020 natural-disaster Royal Commission created a major learning record. Yet recent audits distinguish procedural administration from measured recovery outcomes and continue to find weak monitoring of implementation.

Second, climate and ecological risk impose a hard adequacy constraint. Australia’s first National Climate Risk Assessment explicitly describes cascading, compounding and concurrent effects across infrastructure, supply, health and governance. Observed warming and extremes have intensified. The national adaptation architecture is improving, but its 2025 plan describes the work as a cycle in which prioritising and planning precede implementation and monitoring. The latest national environmental assessment found broad biodiversity decline, insufficient monitoring and ineffective national law in arresting that decline.

Third, ageing is a foreseeable service-and-workforce challenge with unequal geography and burden. The population is projected to reach 31.5 million in 2035–36, with about one quarter of people outside capitals aged 65 or over compared with 17 per cent in capitals. Existing care systems have real scale and areas of progress. They also show unmet need, access barriers and incomplete workforce compliance: 15.4 per cent of older people reported needing more formal assistance in the available national measure; 38.7 per cent of carers reported poor access to formal support; and 45.9 per cent of residential services met both care-minute targets in 2024–25.

Fourth, Australia’s AI-risk baseline is already technological. Business AI adoption rose from 1 per cent in 2022–23 to 12 per cent in 2024–25, while public agencies and cyber systems use automated or AI-enabled tools. This makes a fictional “no technology” baseline invalid. It also makes governance quality urgent. National policy, mandatory APS controls, cyber response and individual agency examples are genuine capabilities. Audits nevertheless show incomplete model assurance, inconsistent benefit measurement and weak comparison with alternatives. Electoral and platform regulators also identify limits in detection, law, transparency and fact-checking support.

Interactions change the judgement

The risks do not arrive in separate departmental trays. Climate events can strain infrastructure, emergency personnel and Defence readiness. Strategic disruption can affect fuel, medicines, cyber systems and care continuity. Ageing changes both demand and workforce availability. AI-driven change can increase cyber and information pressures while depending on electricity, communications, vendors and skilled operators. The national climate assessment directly supports several cross-system links; the Intergenerational Report treats ageing, technology, climate, care demand and geopolitical fragmentation as concurrent long-term forces.

All six risk pairs were screened across P1–P4 and S0–S5. Three are supported and three qualified in the canonical interaction ledger. No unresolved or omitted link carries portfolio weight. The interaction evidence does not assign event probabilities. It asks instead whether a shared function—such as essential supply, care continuity, cyber defence or public information—must continue when more than one pressure is present.

What the strongest non-AI case can still achieve

The strongest rival to any later AI intervention is not inaction. It combines maintained public institutions, staffed services, enforceable rules, independent audit, supply reserves, mutual aid, accessible housing and transport, workforce and migration measures, conventional analytics, open standards, offline fallbacks, public-interest media and structured deliberation. It is sequenced through five checkpoints from 2027 to 2035 and judged by outcomes, not announcements.

This rival is strong enough that Report 2 must beat it function by function. For anticipation and some cyber functions, the non-AI package may meet or partly meet ordinary-condition needs. For climate/ecological continuity, aged-care access, strategic supply and civil preparedness, information integrity and representative deliberation, important gaps remain or are unresolved. Those gaps are specific and testable; they are not a licence for technology-first policy.

The decision this report makes—and the one it does not

This report answers only whether existing systems and best feasible non-AI reforms are adequate to 2035 under the accepted test. Its answer is: overall adequacy is not demonstrated; a defined comparator gap remains.

It hands Report 2 eight exact items: the baseline architecture; evidence gaps; comparator package; constraints; historical effectiveness patterns; distributional vulnerabilities; interaction map; and an unresolved necessity question for each transferred function. Report 2 must then test whether any AI-enabled intervention adds value over that comparator. It may find AI helpful, unnecessary, harmful or unresolved. Report 3 alone owns the final living-well, democratic-renewal, delivery and trilogy verdict.

How to Read This Report

The report proceeds in six numbered passes. Pass 0 states the question and method. Pass 1 fixes the terms and aggregation rules before looking at results. Pass 2 examines the 2000–2025 baseline and the 2026 launch context. Pass 3 constructs the strongest feasible non-AI comparator and judges P1. Pass 4 gives every pathway and condition its strongest good-faith case and sets dated checkpoints. Pass 5 gives the bounded Report 1 answer and exact Report 2 handoff.

Claim markers such as ` resolve to one row in the canonicalevidence-ledger.csv`. That row gives the source ID, exact location, verification state, jurisdiction, limitation or counterevidence and confidence rationale. Appendix A provides readable source links. The other canonical instruments preserve the complete 4,032-cell scorecard, 144-row interaction screen and 3,360-row backcast. The prose is self-contained; the CSVs enable audit rather than carry the argument away from the reader.

Three distinctions prevent overclaiming:

  1. Baseline, comparator and later intervention are different things. P1 includes AI already embedded by 2026, but its reform package does not add an AI intervention for the purpose of this comparison.
  2. Activity is not outcome. A strategy, budget, grant, model, consultation, exercise or adoption count is evidence of intent or activity unless a credible chain demonstrates usable capability and results.
  3. Comparator gap is not AI necessity. A cell that does not meet, partly meets or remains uncertain becomes a question for Report 2, never an automatic case for AI.

Confidence Scale

Rating Meaning in this report
High Direct Australian primary, statutory, operational or audit evidence supports the bounded claim; important limitations are known and stated.
Medium Evidence supports the direction, but transfer, coverage, implementation stage or outcome attribution limits the claim.
Low Evidence is indirect, narrow, old or weakly transferable; it can frame a question but cannot carry a decisive finding.
Insufficient The prescribed search did not locate comparator-grade evidence. The report records UNRESOLVED, an evidence owner and a checkpoint rather than guessing.

Confidence is claim-specific. “High” confidence that a plan exists is not high confidence that the plan works. “Insufficient” is not proof of absence. It identifies what would have to be produced to change the judgement.

The Direct Question, Direct Provisional Answer and Scope

Question: Can Australia’s existing systems and best feasible non-AI reforms adequately manage the interacting risks of war or geopolitical disruption, climate or ecological disruption, population ageing and AI-driven change to 2035?

Direct provisional answer: No—not yet on the accepted evidence and no-averaging rules. Australia has considerable capacity, and the best non-AI package plausibly closes some cells. It is not demonstrated to meet all mandatory outcomes and continuity requirements across the full portfolio. P1 therefore has a comparator gap.

Scope: The report examines Australia’s 2000–2025 experience, the 2026 launch baseline and feasible reform checkpoints through 2035. It tests anticipation, prevention, response, recovery, learning and implementation; six transferred functions; seven domains; seven distribution groups; four pathways; and six operating conditions. It does not estimate the probability of wars, disasters, demographic paths or AI events. It does not decide whether AI is necessary, useful or net beneficial. It does not determine living-well or democratic-renewal causality, final delivery feasibility, or the trilogy’s final hypothesis.

Pass 0 — The Question and Method

0.1 The proposition being tested

The proposition is deliberately demanding: Australia’s launch-baseline systems plus the strongest feasible non-AI reforms meet the accepted adequacy test across the interacting four-risk portfolio, seven domains and S0–S5 to 2035. It can pass, fail in defined cells, or remain unresolved. The method does not begin with the conclusion that institutions are failing, nor with the conclusion that AI is a solution.

The unit of analysis is a risk–function cell. The four risks are:

  • RWAR: war and geopolitical disruption, including strategic supply, civil preparedness, cyber and information operations;
  • RCLM: climate and ecological disruption, including compound hazards, ecosystem decline and infrastructure stress;
  • RAGE: population ageing, including care demand, workforce, fiscal and geographic effects; and
  • RAIC: AI-driven change, including adoption, administrative automation, cyber, labour and information-system effects.

The six transferable functions are anticipation and forecasting (FANT); climate, infrastructure and logistics continuity (FCIL); care and living-well support (FCAR); cyber and AI-threat management (FCYB); information judgement (FINF); and inclusive deliberation (FDEL). These are capabilities, not technologies. A function can be supplied through people, institutions, material infrastructure, conventional software, law, markets, community practice or a combination.

0.2 The historical test

For each risk, the report asks what Australia’s systems did from 2000 to 2025 in six capacities: anticipation, prevention, response, recovery, learning and implementation. This avoids a familiar distortion: pointing to an inquiry as proof of learning, to spending as proof of recovery, or to a strategy as proof of implementation. Twenty-four structural cells are assessed in Pass 2.

The same outcome-and-continuity rule applies to positive and negative evidence. A system counts as adequate when it delivers the required function, maintains it through the relevant condition, protects each mandatory domain and distribution group, learns from failure and implements corrections on time. Evidence of inadequacy must address the same object: the absence, interruption, unequal distribution or non-implementation of that function. Inputs and activities are not silently compared with outcomes.

0.3 Evidence hierarchy and rival

The build preferred Australian legislation, official statistics, statutory inquiries, national assessments, operational reports and performance audits. Peer-reviewed or authoritative international material could be used only with an explicit Australian transfer test. Repository reports supplied framing, method and source leads, not inherited empirical proof. Method and provenance sources are Class E and carry no empirical weight.

Every load-bearing claim has a stable ID. Its ledger row contains an exact location rather than “whole report”; a current verification state; counterevidence or a limitation; jurisdiction and transfer note; and a written confidence reason. Reproducible absence findings name the search record, owner and next checkpoint. This is important because “not publicly located” can mean protected, unpublished, unevaluated or genuinely absent. The report does not collapse those possibilities.

The strongest rival is the best feasible non-AI reform package, not the unreformed status quo. It includes the institutional competence, workforce, public services, social protection, regulation, standards, conventional digital and rules-based systems, material capacity, redundancy, maintenance, procurement, mutual aid and trusted democratic practice that Australia could plausibly deliver by 2035. That package receives owners and sequencing before being judged.

0.4 The four pathways and six conditions

The accepted pathways are:

  • P1: the 2026 launch baseline, including already embedded AI, plus strongest feasible non-AI reform;
  • P2: P1 plus bounded assistive AI, to be tested only in Report 2;
  • P3: P1 plus coordinated public-interest AI capability, to be tested only in Report 2; and
  • P4: P1 plus deeper AI-enabled institutional transformation, to be tested only in Report 2 and subject to the same gates.

The accepted conditions are:

  • S0: ordinary but changing conditions;
  • S1: war or severe geopolitical and supply disruption;
  • S2: severe climate or ecological disruption;
  • S3: acute ageing, care and workforce stress;
  • S4: democratic and information-system stress; and
  • S5: severe AI or cyber disruption.

They are discriminating conditions, not forecasts and not probability claims. P2–P4 receive their strongest good-faith conceptual case in Pass 4, but their outcomes remain not-yet-tested here.

0.5 What counts as an answer

At cell level, P1 can meet, partly meet, not meet, be uncertain or be UNRESOLVED. At portfolio level, “adequate” requires all mandatory domains and gates to pass for the functions made portfolio-critical by supported or qualified interactions. An unresolved interaction cannot make a function critical; an unresolved critical function cannot be treated as a pass.

What this means

The method makes this a harder question than whether Australia has plans, agencies or money. It asks whether functions work, continue under stress, reach people fairly, avoid ecological and democratic failure and improve when evidence exposes a weakness. It also protects the later AI inquiry from a weak comparator: Report 2 must compete against serious conventional reform, not against neglect.

Pass 1 — Terms, Rules and Aggregation

1.1 Seven non-substitutable domains

Code Domain Outcome and continuity question
D01 Peace, security and sovereignty Can Australia protect people, lawful authority, essential capability and strategic agency without creating unacceptable dependency?
D02 Ecological integrity and climate safety Does action reduce exposure and preserve ecological systems without shifting lifecycle or territorial harm elsewhere?
D03 Health, care and human capability Can people obtain timely, dignified care and maintain functional capability, relationships and agency?
D04 Democracy and information integrity Can people access evidence, judge claims, contest decisions, hear minority reasons and exercise legitimate influence?
D05 Fair livelihoods and economic security Are work, income, time, adjustment costs and benefits fairly distributed, including for carers and people outside secure employment?
D06 Infrastructure, place and essential services Do energy, water, communications, transport, housing, food, medicine and local services continue and recover across geography?
D07 Institutional capability, learning and implementation Are owners authorised, staffed and accountable; are plans implemented; and do monitoring and review change practice?

Each scorecard row points to its domain method (AHL1-MTH-D01 to AHL1-MTH-D07). These methods are not empirical sources. They preserve a common test across favourable and unfavourable evidence.

1.2 No averaging and vetoes

National resilience is not a single score. A strong cyber response cannot compensate for ecological collapse. High average access cannot compensate for an inaccessible service imposed on disabled people. A fast response cannot compensate for a recovery system that leaves regional communities without continuity. For that reason:

  • ecological failure in D02 is a mandatory veto;
  • democratic or information failure in D04 is a mandatory veto;
  • unacceptable security, sovereignty, human-agency or fair-distribution failure is a mandatory gate;
  • harm exported to another place, group, generation or supply chain remains harm; and
  • a material group result cannot be replaced by a national average.

The ecological veto is especially consequential. Australia’s national environment assessment found that most biodiversity indicators were declining, information remained incomplete, and the principal national law had not arrested decline. Later reform may improve that position, but an announced reform cannot erase the measured baseline.

The democratic and information veto does not demand unanimity or perfect trust. It demands lawful decisions, accessible evidence, reasons, contestability, representative opportunity to influence, and correction when systems mislead or harm. Robodebt demonstrates why a technically administered program cannot be counted adequate when legality, reasons, review and human consequences fail. The Royal Commission’s 57 recommendations are a powerful learning record, but consultation on an automated-decision framework remains an implementation step rather than completed protection.

1.3 Distribution is part of capability

Seven group lenses apply wherever evidence permits: age, disability, income, geography, gender, employment and caring status. These often overlap. In 2022, 21.4 per cent of Australians had disability, 43.8 per cent of primary carers had disability, and primary carers were concentrated in lower-income households. Regional ageing also differs markedly from capital-city ageing. These are not side notes: they change demand, accessibility, workforce and recovery requirements.

The canonical scorecard records all seven lenses in every row. Where a cell lacks group-specific outcome evidence, it says so and assigns the gap to a 2027 checkpoint. A non-digital channel, accessible format or remote service is not an optional kindness; it is part of continuity. The Australian Digital Inclusion Index provides corroborating evidence of persistent affordability and inclusion gaps for older, low-income, disabled and regional people, but it is Class F evidence and does not carry a national adequacy verdict by itself.

1.4 Outcome and continuity grammar

The evidence is classified along a delivery chain:

Evidence object What it can establish What it cannot establish alone
Announcement or strategy intent, priority, authority implemented capacity or outcome
Appropriation or spending resources committed or paid value, access, continuity or result
Activity or exercise an operation occurred sustained real-world performance
Adoption a tool or practice is present benefit, resilience, necessity or fairness
Implemented capability a usable function exists universal access or performance under every condition
Measured outcome observed result in a defined setting automatic transfer to other groups, places or conditions

This is why the 2026 Defence strategy strengthens the baseline but does not prove civil preparedness; why a disaster fund award does not prove mitigation before projects are implemented; and why 12 per cent business AI use establishes embedded technology but not benefit.

1.5 Portfolio aggregation

Interaction status is one of supported, qualified, unresolved or omitted. Supported and qualified links can make their affected functions portfolio-critical. Qualified means the mechanism is sufficiently evidenced to matter, but its magnitude, condition or transfer remains bounded. Unresolved and omitted links remain visible and carry no aggregation weight.

Once eligible, a function is judged across its affected domains and conditions without a weighted average. A critical function passes only when its outcome, continuity, distribution and mandatory gates pass. The portfolio passes only when every critical function passes. This conservative rule prevents a large number of low-stakes green cells from hiding one catastrophic red cell.

What this means

The report does not treat resilience as a league table. It treats it as a set of promises that must hold together: essential services continue, people can participate and contest, ecological systems are not sacrificed, and vulnerable groups are not used as shock absorbers. A single unresolved critical promise blocks an overall “adequate” finding but does not prove that a particular remedy will work.

Pass 2 — Baseline and Interactions

2.1 The 2026 launch context

Australia begins this inquiry with strong public institutions by comparative standards, a federal system able to mobilise large resources, mature scientific and statistical agencies, professional emergency and defence organisations, extensive care and social-protection systems, electoral integrity institutions, and significant community capability. The OECD’s Australian survey found 46 per cent with high or moderately high trust in the federal government and the same share feeling they had a say, with Australia above OECD averages on several measures. That is a foundation, not a universal pass: voice, gender and financial-security gaps were large.

The launch context is also one of concurrent change. Treasury’s Intergenerational Report frames population ageing, technological transformation, climate change, rising care demand and geopolitical fragmentation as connected long-term forces. It also records that previous projection assumptions can be overtaken by behaviour and policy—an argument for adaptive checkpoints, not false precision.

AI is already inside P1. Businesses, the ATO, IP Australia, cyber agencies and other public bodies use automated or AI-enabled systems. The National AI Plan and the updated responsible-AI policy for government form part of the policy baseline. Removing that technology in imagination would misstate current systems. The comparison instead asks what the baseline plus non-AI institutional and material reforms can deliver, while holding new AI additions for Report 2.

2.2 The 24-cell historical baseline

The table below compresses the 4 × 6 baseline. “Material” means a functioning capacity is evidenced; “partial” means important capacity and an outcome or implementation gap coexist; UNRESOLVED means the prescribed public search did not locate a comparator-grade result. The status is not an agency score.

Risk Anticipation Prevention Response Recovery Learning Implementation
RWAR Material: current strategy and risk monitoring identify sovereignty, supply and civil-preparedness needs. Partial: cyber hardening and supply work exist; Defence inventory risk treatment was not operationalised in the audited case. Material/partial: established Defence, Home Affairs and ACSC response; activity and exercises do not prove all-service continuity. UNRESOLVED: no open national S1 civilian recovery outcome test located. Partial: inquiries, audits and exercises generate lessons; public closure evidence varies. Partial: major-project schedule slippage and incomplete supply implementation constrain delivery.
RCLM Material: national science, forecasting and the 2025 risk assessment identify compound risks. Partial/not meet ecologically: prevention is a stated priority, while biodiversity decline and grant-stage delivery gaps remain. Material: mature multi-level emergency response and mutual aid, with concurrency constraints. Partial: large recovery funding and largely effective administration coexist with weak outcome monitoring. Material/partial: the 2020 Royal Commission and repeated reviews provide learning, but closure must be evidenced. Partial: a national adaptation cycle exists, while implementation, monitoring and policy contribution remain incomplete.
RAGE Material: current population projections reveal national and regional ageing. Partial: home support and workforce reforms can delay avoidable loss, but demand and access are not demonstrated met. Material/partial: large care systems respond daily, with unmet need and workforce constraints. Partial: ongoing care and social protection provide continuity, but group and place outcomes remain uneven. Partial: quality data, audits and reform reviews exist; evaluation planning is not consistently embedded. Partial/not meet: only 45.9% of residential services met both care-minute targets and CHSP objectives/demand were not demonstrated.
RAIC Material/partial: adoption statistics, cyber reporting and agency portfolios make change visible; outcome measurement remains thin. Partial: mandatory APS governance and cyber controls exist, with assurance gaps. Material: ACSC incident response and public-administration review institutions are active. Partial/uncertain: continuity plans are widespread but not equivalent to tested recovery; human redress pathways remain system-specific. Material/partial: Robodebt and ANAO audits create strong lessons; implementation and alternative comparison are incomplete. Partial: policy adoption and agency tools are real, but model assurance, evaluation and information governance remain incomplete.

2.3 What the baseline says about each risk

War and geopolitical disruption

The strongest positive case is institutional depth. The 2026 National Defence Strategy explicitly prioritises self-reliance, sovereign industry, diversified supply and civil preparedness. The Office of Supply Chain Resilience monitors vulnerabilities in essential goods, including fuel. ASD’s ACSC responded to more than 1,200 cyber incidents in 2024–25 and ran 17 exercises involving more than 120 organisations. These are real anticipation and response capacities.

The limiting evidence concerns delivery and public continuity proof. Across 21 major Defence projects at June 2025, average and median schedule slippage were 21 months, although reported scope confidence was high and eight projects had no total slippage. A separate audit recorded that a general-stores supply-chain risk solution had not yet been operationalised. Public information establishes a resilience office’s remit but not a nationally tested essential-goods outcome under S1. The full 2025 Critical Infrastructure Annual Risk Review was not publicly located; only its official release supports a qualified interaction finding.

This does not mean protected capability is absent. It means a public report cannot convert protected or unlocated evidence into a pass. The checkpoint response is a publishable assurance case: protected details can remain protected while an independent body reports whether fuel, medicine, communications and civilian recovery thresholds were tested and met.

Climate and ecological disruption

Australia’s anticipation is much stronger than a decade ago. The National Climate Risk Assessment identifies cross-system risks and explicitly analyses cascades and concurrency. State of the Climate 2024 reports 1.51 ± 0.23°C warming since 1910 and intensified extremes. Emergency management is supported by national, state and local systems and by a substantial history of response.

The decisive weakness is the step from knowledge and funding to prevention, ecological outcomes and sustained implementation. The 2025 National Adaptation Plan establishes roles and a useful cycle, but describes the current national step as “prioritise and plan”, with adaptation, response and monitoring continuing. An ANAO climate-governance audit found key plans and frameworks not delivered and could not demonstrate the contribution of particular policies to the overall reduction. Disaster recovery provides large financial response, yet audited monitoring did not demonstrate performance against the arrangement principles.

Ecology prevents a portfolio pass at launch. The State of the Environment finding is not that every species or intervention failed; it records some targeted success. Its national conclusion is nevertheless broad decline, cumulative pressure, insufficient monitoring and ineffective law in arresting decline. Under the ecological veto, improvement elsewhere cannot average that away.

Population ageing

Ageing is predictable enough to plan for but not uniform enough for a national average to solve. The 2025 Population Statement projects a population of 31.5 million in 2035–36 and a median age near 40 over the next decade. It also projects a much larger 65-plus share outside capitals. Australia’s working-age population is still projected to grow, which is positive counterevidence to deterministic decline. Migration, participation, health, housing and care policy can change the outcome.

Existing systems have reach. More than 800,000 people received Commonwealth Home Support Program services, surveyed clients were largely satisfied, and the share of residential services meeting both care-minute targets improved from 34.0 to 45.9 per cent. The strongest case for P1 therefore includes expansion and redesign of institutions already operating, not their wholesale replacement.

Yet the adequacy test is unmet where assessed need, access, workforce and distribution are not demonstrably secured. Available national measures found unmet demand among older people and poor formal-support access among carers. The CHSP audit could not demonstrate that demand or program objectives were met. Disability, carer and income overlap means a digital-only or family-substitution response could export public-system pressure into unpaid households.

AI-driven change

The baseline contains both capability and warning. AI adoption is measurable, and the National AI Plan supplies national intent. The DTA’s updated policy requires foundational training and governance in Commonwealth entities. IP Australia provides a positive narrow example: its use was judged largely effective. Cyber response, electoral education and public-sector audits provide institutions able to notice and address harms.

The warning is that adoption and governance paperwork are ahead of demonstrated outcomes. ANAO found the ATO’s arrangements partly effective and reported that 74 per cent of production models lacked completed data-ethics assessments at August 2024. At IP Australia, benefits were inconsistently measured, alternatives were not systematically documented and monitoring was partly effective. The AEC warns there is no universal detection mechanism and that deepfakes are not generally prohibited by electoral law; ACMA found only modest improvement in platform reporting, stalling fact-checking support and a voluntary code.

The relevant non-AI functions are regulation, assurance, education, contestability, public information, cyber hygiene, procurement discipline and human service continuity. One search did not locate a national outcome evaluation showing that embedded AI in aged care improves continuity and distribution over the strongest workforce and community-care comparator. That finding is UNRESOLVED, not negative proof and not a Report 1 AI judgement.

2.4 All six risk interactions

Pair Status Shared mechanism Portfolio-critical functions
RWAR–RCLM Supported Compound pressure on infrastructure, supply, responders and Defence readiness. The climate assessment identifies concurrency that can overwhelm domestic response and increase reliance on Defence. FANT, FCIL, FCAR
RWAR–RAGE Qualified Strategic disruption raises medicine, fuel, care-workforce and fiscal pressure in an older, regionally uneven population. Exact S1 continuity remains untested. FCIL, FCAR, FANT
RWAR–RAIC Supported Geopolitical actors, cyber compromise, concentrated vendors and AI-enabled information operations can disrupt critical services; Australian cyber operations demonstrate the live coupling. FCYB, FINF, FCIL, FANT
RCLM–RAGE Supported Heat, smoke, displacement and outages increase health and care demand while older, disabled and caring populations face unequal access. FCAR, FCIL, FANT
RCLM–RAIC Qualified Climate outages threaten data-dependent services; digital infrastructure adds energy, supply and cyber dependencies. The AI-specific lifecycle increment remains for later testing. FCIL, FCYB, FANT, FINF
RAGE–RAIC Qualified Ageing raises care and workforce demand while digitisation changes service access, work and information, with exclusion risks. AI care benefit remains unresolved. FCAR, FINF, FDEL, FANT

The canonical ledger repeats the screen for every pathway and condition because eligibility can persist while mechanisms and evidence requirements change. Only supported and qualified rows affect aggregation. No rhetorical seventh interaction has been introduced, and no event likelihood has been assigned.

2.5 Changed conditions to 2035

The 2035 frame is not “more of today” and not a prediction of a single future. It contains observed climate trends, conditional demographic projections, strategic-policy changes, rising digital adoption and new governance arrangements. The relevant question is whether systems can learn and maintain functions as those conditions change.

Several changes are already locked into planning needs: a larger and older population; stronger regional ageing; more AI embedded in markets and administration than in 2022; climate exposure acting through infrastructure and service networks; and a defence posture seeking greater self-reliance. Others remain uncertain in timing and magnitude. The report handles uncertainty with S0–S5 stress conditions and dated evidence checkpoints, not event probabilities.

What this means

Australia’s baseline is neither broken nor adequate by default. It is a capable but uneven system: better at seeing danger and mobilising response than at proving prevention, fair continuity, recovery outcomes and completed learning. Interactions make those weak links more important because the same workforce, infrastructure, information and institutional attention must serve several risks at once.

Pass 3 — Comparators and P1 Sufficiency

3.1 What “strongest feasible” means

The comparator assumes serious implementation. It does not assume unlimited money, perfect coordination or instant legislation. It does assume that governments use established authorities, fund priority capabilities, maintain skilled workforces, enforce adopted rules, publish outcome measures and correct failures over a nine-year period. Each package follows a common sequence: authority and design by 2027; operational capability by 2029; measured outcomes by 2031; cross-condition stress test by 2033; and sustained result by 2035.

Its recurring components are:

  • independent risk observatories and transparent assumptions;
  • prevention investment tied to measured exposure reduction;
  • public-service and care workforces with regional and accessibility design;
  • strategic reserves, diversified procurement, interoperability and offline fallback;
  • enforceable safety, automated-decision, cyber and information rules;
  • conventional analytics and rules-based systems where they are sufficient;
  • representative deliberation with a published government response duty;
  • independent audit, outcome dashboards and mandatory correction triggers; and
  • social protection and transition support that do not export cost to carers or insecure workers.

3.2 The 24 strongest feasible comparator packages

Risk Function and comparator Strongest non-AI package 2035 P1 status and reason
RWAR FANT — AHL1-CMP-RWAR-FANT Defence–Home Affairs horizon scanning; independent assumptions; public risk summaries; exercises; parliamentary and audit follow-up. Partly meets: institutions and strategy are strong; full critical-infrastructure evidence and civilian stress outcomes are incomplete.
RWAR FCIL — AHL1-CMP-RWAR-FCIL Essential-goods mapping; diversified suppliers; fuel, medicine and spares assurance; logistics interoperability; protected stock tests; civilian recovery benchmarks. Does not meet/UNRESOLVED: supply monitoring exists, but audited implementation gaps and no open national S1 recovery outcome prevent a pass.
RWAR FCAR — AHL1-CMP-RWAR-FCAR Health and care continuity standards; priority medicine stocks; surge workforce; accessible local plans; carer relief and non-digital contact. Uncertain: relevant care and continuity systems exist, but combined S1 group outcomes are not publicly demonstrated.
RWAR FCYB — AHL1-CMP-RWAR-FCYB Essential Eight enforcement; segmented/offline operation; cross-sector exercises; supplier assurance; incident recovery metrics and public aggregate reporting. Partly meets: material response and exercise capacity; broader continuity and full interaction detail remain incomplete.
RWAR FINF — AHL1-CMP-RWAR-FINF Trusted official channels; rapid evidence correction; independent public media; provenance rules; accessible multilingual communication; rights-preserving emergency powers. Partly meets: mature institutions and campaigns exist; detection, law and platform accountability have gaps.
RWAR FDEL — AHL1-CMP-RWAR-FDEL National civil-preparedness assemblies and local panels; representative recruitment; protected minority reasons; formal government response; parliamentary scrutiny. UNRESOLVED: no national evaluated mechanism meeting the response-duty and influence test was located.
RCLM FANT — AHL1-CMP-RCLM-FANT Maintain BOM/CSIRO and Australian Climate Service capability; interoperable hazard, exposure and vulnerability data; local translation and triggers. Meets ordinary anticipation/partly meets portfolio: high-quality national assessment exists; translation into delivered adaptation is incomplete.
RCLM FCIL — AHL1-CMP-RCLM-FCIL Mandatory asset-risk disclosure; maintenance; redundancy; land-use reform; resilience standards; prevention grants tied to outcomes; recovery metrics. Does not meet: response and funding capacity are strong, but prevention delivery, recovery outcomes and concurrency remain insufficiently demonstrated.
RCLM FCAR — AHL1-CMP-RCLM-FCAR Accessible heat, smoke and evacuation plans; primary-care surge; resilient housing; community networks; ecological restoration as protective infrastructure. Does not meet: the ecological veto is engaged at baseline and unequal care continuity remains material.
RCLM FCYB — AHL1-CMP-RCLM-FCYB Resilient power and communications; offline modes; tested backup; cyber-safe restoration sequencing; independent exercises. Partly meets: emergency and cyber institutions are substantial, but compound outage outcomes are not nationally demonstrated.
RCLM FINF — AHL1-CMP-RCLM-FINF Authoritative warnings; accessible risk maps; uncertainty communication; local trusted intermediaries; open correction and misinformation response. Partly meets: national science and warnings are strong; reach, trust and influence differ by group and place.
RCLM FDEL — AHL1-CMP-RCLM-FDEL Place-based adaptation panels; First Nations and affected-community authority; published trade-offs; appeal; formal response and monitoring. Uncertain: established consultation is not equivalent to representative influence and measured implementation.
RAGE FANT — AHL1-CMP-RAGE-FANT Rolling demographic, care-demand and workforce projections; local service maps; scenario ranges; annual actual-versus-plan review. Meets anticipation/partly meets implementation: high-quality projections exist; service translation remains uneven.
RAGE FCIL — AHL1-CMP-RAGE-FCIL Age-friendly housing, transport and local services; accessible digital and non-digital channels; regional workforce housing and service guarantees. Partly meets: mature infrastructure and programs exist, but regional and inclusion gaps remain.
RAGE FCAR — AHL1-CMP-RAGE-FCAR Pay and train care workers; expand home/community care; support carers; integrate health, disability and aged care; publish need, wait and outcome measures. Does not meet: unmet need, carer access and workforce-target evidence breach continuity and distribution requirements.
RAGE FCYB — AHL1-CMP-RAGE-FCYB Secure service portals; identity and fraud protection; staffed recovery; accessible alternatives; provider cyber standards and drills. Partly meets: national cyber capability is real, but care-provider and group-specific continuity evidence is incomplete.
RAGE FINF — AHL1-CMP-RAGE-FINF Plain-language advice; navigators; non-digital access; accessible formats; scam education; trusted community intermediaries and appeals. Partly meets: existing channels help, while affordability, disability and regional gaps remain.
RAGE FDEL — AHL1-CMP-RAGE-FDEL Older people, disabled people, carers and care workers in representative design panels with published responses and paid participation. Uncertain: survey and consultation evidence does not yet show sustained decision influence and group outcomes.
RAIC FANT — AHL1-CMP-RAIC-FANT Independent technology observatory; incident and adoption statistics; procurement registers; conventional modelling; labour and distribution monitoring. Partly meets: adoption and agency evidence exist, but benefits and alternatives are inconsistently measured.
RAIC FCIL — AHL1-CMP-RAIC-FCIL Vendor concentration limits; open standards; data portability; conventional fallback; resilient compute, energy and communications; exit tests. Uncertain: adjacent cyber and climate evidence supports the need; national outcome-grade concentration and recovery tests are incomplete.
RAIC FCAR — AHL1-CMP-RAIC-FCAR Workforce-first care; accessible assistive technology without new AI; relationship and contact floors; community care; human review and non-digital access. Partly meets comparator, AI comparison unresolved: conventional reforms are credible but current care outcomes fail; no national AI-over-comparator evaluation was located.
RAIC FCYB — AHL1-CMP-RAIC-FCYB Mandatory assurance, red teams, inventories, training, incident reporting, procurement controls, cyber hygiene and accountable human authority. Partly meets: controls and response exist; ATO assurance gaps and self-reported preparedness prevent a portfolio pass.
RAIC FINF — AHL1-CMP-RAIC-FINF Enforceable platform transparency; public-interest media; provenance; education; reasons, human review and appeal in automated administration. Does not meet/partly implemented: strong learning and reform activity coexist with voluntary platform rules and incomplete legal implementation.
RAIC FDEL — AHL1-CMP-RAIC-FDEL Citizens’ panels and parliamentary processes on high-impact deployment; representative recruitment; expert contest; minority reports; formal response. Uncertain: institutional foundations exist, but national evaluated influence and distribution evidence are incomplete.

3.3 P1 judgement

P1’s strongest case is credible. By 2035, conventional reform could improve risk translation, supply assurance, workforce access, cyber recovery, public information and deliberation. Australia has institutions capable of doing that work. The result is not a blanket failure.

P1 nevertheless does not meet the portfolio adequacy test on current evidence. Three kinds of cell block the pass:

  1. Observed outcome failures, especially ecological decline, care access and workforce-target gaps.
  2. Implementation gaps, where strategy, funding or policy has not yet become measured capability, including adaptation, mitigation grants, automated-decision safeguards and some supply assurance.
  3. Critical unresolved evidence, including national civilian S1 recovery and an evaluated national deliberative mechanism for civil preparedness.

No averaging means “mostly capable” is not “adequate”. The ecological veto is already engaged. D03 and D06 distribution and continuity failures remain material. D04 information and influence evidence is incomplete. S1–S5 cross-condition continuity is not fully demonstrated.

This is the report’s central distinction: P1 has a comparator gap; the evidence does not establish an AI-shaped hole. This is not a finding that AI is necessary, and it is not a finding that AI is not necessary. Some gaps may be best closed by funding and enforcing P1. Some may be inherently political or distributive. Some may be inaccessible to public evidence. Report 2 must test each transferred function against this strengthened package and is free to conclude that no additional AI intervention is warranted.

What this means

The best conventional reform package deserves to be treated as a real competitor. It can plausibly solve several weaknesses and should proceed where justified regardless of later AI choices. But a plausible package is not a proven outcome. Until ecological, care, continuity, information and implementation thresholds are met, Australia cannot claim full portfolio adequacy—and no one can claim that AI is the missing ingredient merely because P1 falls short.

Pass 4 — Pathways, Conditions and Backcast

4.1 Strongest good-faith case for P1–P4

Pathway Strongest good-faith case State after Report 1
P1 Existing institutions and embedded 2026 technology, strengthened through the complete non-AI comparator, may deliver adequate functions without additional AI complexity or dependency. Tested here: overall does not meet/is not demonstrated; cell findings range from meets to UNRESOLVED.
P2 Bounded assistive AI could help workers retrieve, translate or analyse information while humans retain authority and non-AI fallback. Not yet tested: Report 2 must show additionality over P1, not convenience or adoption.
P3 Coordinated public-interest AI capability could overcome fragmentation, share assurance and serve functions markets underprovide. Not yet tested: Report 2 must test sovereignty, distribution, ecology, governance and institutional alternatives.
P4 Deeper transformation could redesign anticipation, services and deliberation where incremental tools cannot meet interacting demands. Not yet tested: the burden is highest because dependency, lock-in, agency and systemic harm increase with transformation.

The P2–P4 descriptions are not endorsements. They prevent a straw-person comparison. The canonical scorecard labels every later intervention and AI outcome not-yet-tested — Report 2 owner.

4.2 How S0–S5 discriminate

S0 asks whether reform works in ordinary but changing conditions, including implementation discipline. S1 tests essential supply, lawful authority, civilian continuity and recovery during severe strategic disruption. S2 tests compound climate and ecological disruption, including simultaneous regional demands. S3 tests care, health, income and workforce continuity under acute ageing pressure. S4 tests information judgement, contestability and representative influence during democratic stress. S5 tests prevention, response, recovery and non-digital continuity during severe AI or cyber disruption.

The strongest P1 package changes emphasis by condition without changing its principles. Under S1 it relies on stocks, diversification, exercises and trusted communication. Under S2 it relies on prevention, resilient assets, local capacity and ecological restoration. Under S3 it relies on staffed care, accessible housing and social protection. Under S4 it relies on independent media, transparent reasons, law and deliberation. Under S5 it relies on secure design, segmentation, manual fallback, accountable authority and rapid recovery.

4.3 The dated backcast

The canonical backcast contains one row for every checkpoint, pathway, risk, function and domain—3,360 rows. The table below gives the self-contained decision spine.

Checkpoint Required evidence Failure trigger Required response and principal owners
2027-12-31 — authority and baseline Every critical function has a named owner, public baseline, domain method, group breakdown where material, budget/workforce path and independent assurance plan. Full or independently summarised critical-infrastructure evidence is available. No accountable owner; no baseline; plans treat adoption/spending as outcomes; a material group is unmeasured without an evidence owner. Escalate to Cabinet and responsible ministers; legislate or assign authority; fund ABS/portfolio data; establish independent assurance. Owners: PM&C, Finance, Treasury, portfolio departments, states and territories.
2029-12-31 — operational capability Comparator packages are in operation: prevention projects delivered, essential stocks and fallbacks tested, care access capacity expanded, automated-decision and platform safeguards enforceable, representative panels established. Milestones slip without recovery plan; critical assets or services cannot demonstrate exercises; ecological or democratic veto indicators deteriorate. Reallocate from announcements to delivery; strengthen enforcement; add workforce and local capacity; publish correction plan. Owners: NEMA, Defence, DCCEEW, Health, Home Affairs, DTA, AGD, ACMA and states.
2031-12-31 — measured outcomes Outcome data show reduced exposure and recovery time, care need and access improving, workforce standards sustained, reasons and appeals working, distribution gaps narrowing and no exported ecological harm. Only inputs/adoption are reported; a critical function remains UNRESOLVED; group or place outcomes worsen; independent evaluation cannot attribute improvement. Commission evaluation; pause or redesign failing measures; use non-digital and community fallbacks; increase prevention or service capacity. Owners: ANAO and evaluators with accountable portfolios.
2033-12-31 — cross-condition stress test Multi-risk exercises test S1–S5, including simultaneous climate, supply, care, cyber and information demands; recovery and learning actions close on schedule. One system’s response disables another; Defence becomes the unplanned default for civilian concurrency; manual fallback fails; lessons remain open. Invoke cross-portfolio remediation; establish surge agreements and protected reserves; repeat the test after correction. Owners: NEMA, National Security Committee portfolios, states, critical operators and care systems.
2035-12-31 — sustained adequacy All portfolio-critical functions meet outcomes and continuity across domains and groups; ecological and democratic/information vetoes pass; independent evidence shows reforms sustained. Any critical cell does not meet or is unresolved; any veto fails; harm is exported; a later AI pathway lacks additionality over P1. Do not claim portfolio adequacy. Retain or strengthen the best-performing comparator, publish the failure and return the affected function to bounded redesign and review. Owners: Parliament, Cabinet, auditors and portfolio owners.

4.4 Function-specific discriminators

The backcast is deliberately falsifiable. For FANT, publication alone is not success: owners must show that forecasts changed a decision before harm. For FCIL, a plan is not enough: service restoration, stock endurance and geographic reach must be observed. For FCAR, places or minutes are not complete outcomes: access, function, dignity, continuity, human contact and carer burden matter. For FCYB, plan possession is not recovery: exercises and incidents must show essential functions continuing. For FINF, impressions and content volume are not judgement: reasons, correction, reach, appeal and decision quality matter. For FDEL, consultation attendance is not influence: recruitment, minority reasons, a formal response and traceable decision change are required.

4.5 Owners of unresolved evidence

Three absence findings are important enough to carry named ownership. Home Affairs, Defence and the Office of Supply Chain Resilience own the public assurance gap for S1 civilian service recovery. Home Affairs and Parliament own the national civil-preparedness deliberation evidence gap. Health owns a comparator-grade evaluation of AI already embedded in aged care, though the AI additionality judgement belongs to Report 2. The full Critical Infrastructure Annual Risk Review should be produced publicly or replaced by an independently assured summary that reveals no protected operational detail.

What this means

The pathway choice does not have to be made in 2026 by assertion. Australia can demand evidence at five dates and act when a threshold fails. The checkpoints turn “be more resilient” into accountable questions: who owns the capability, what continued, who missed out, what harm was displaced, what changed after learning, and whether any later AI addition actually beats the strongest conventional reform.

Pass 5 — Report 1 Answer and Exact Report 2 Handoff

5.1 Answers to the four Report 1 tests

R1-T1 — What can existing systems do? They provide substantial anticipation, response and institutional learning. Australia has high-quality climate and demographic science, mature emergency and cyber response, extensive care systems, regulatory and audit capacity, and a stable democratic base. Capability is real, not hypothetical.

R1-T2 — Where do historical patterns constrain confidence? Prevention, recovery-outcome monitoring, learning closure and implementation repeatedly lag plans and response. Defence delivery and supply assurance, climate adaptation and ecology, care access and workforce targets, AI assurance and information governance contain material gaps.

R1-T3 — Can the strongest non-AI package meet the accepted test? It can plausibly meet some function-condition cells and materially improve many others. The report cannot demonstrate that it meets the full portfolio by 2035. Ecological outcomes, care and distribution, multi-risk continuity, democratic/information safeguards and unresolved critical evidence block an overall pass.

R1-T4 — What follows? Only a bounded Report 2 test. Each comparator gap becomes a necessity question: is an additional AI intervention required for this function, does it add outcome and continuity value over P1, and can it pass every mandatory gate? No affirmative answer is inherited.

5.2 Direct Report 1 verdict

Australia’s existing systems and strongest feasible non-AI reforms are not demonstrated adequate across the interacting four-risk portfolio to 2035. P1 therefore contains a provisional comparator gap.

This verdict is not “Australia cannot cope”, not “non-AI reform is futile”, and not “AI is necessary”. It is a strict portfolio judgement under no averaging and veto rules. Several P1 cells meet or partly meet. Others do not meet, are uncertain or remain UNRESOLVED. The strongest rival remains live and must be funded, implemented and evaluated rather than held artificially weak.

5.3 Exact eight-item handoff to Report 2

Handoff ID Exact Report 1 OUT / Report 2 IN Transferred content
AHL1-OUT-01 Baseline architecture Four risks × six historical capacities; the 24 risk–function cells; seven domains; 2026 embedded-AI state; current owners and evidence.
AHL1-OUT-02 Evidence gaps UNRESOLVED full critical-infrastructure review, S1 civilian recovery, national civil-preparedness deliberation and aged-care AI-over-comparator outcomes, with search records, owners and checkpoints.
AHL1-OUT-03 Strongest feasible non-AI comparator package Exactly 24 named comparator packages, common 2027–2035 sequence, constraints and P1 cell status.
AHL1-OUT-04 Constraints No averaging; ecological and democratic/information vetoes; distribution, agency, security/sovereignty and no-exported-harm gates; implementation and continuity requirements.
AHL1-OUT-05 Historical effectiveness patterns Relative strength in anticipation/response; recurrent prevention, recovery-outcome, learning-closure and implementation gaps; plans/spending/activity/adoption separated from capability/outcomes.
AHL1-OUT-06 Distributional vulnerabilities Age, disability, income, geography, gender, employment and caring lenses, including overlapping disability/carer/income and regional-ageing evidence.
AHL1-OUT-07 Interaction map Six screened risk pairs across P1–P4/S0–S5; only supported/qualified links eligible for portfolio aggregation; affected functions and domains preserved in the canonical ledger.
AHL1-OUT-08 Unresolved necessity question by transferred function For each comparator gap: can a specified AI intervention deliver material additional outcome and continuity value over P1, within time and constraints, without failing any mandatory gate?

This handoff is reciprocal with the locked Report 2 boundary. Report 2 must not rewrite the P1 comparator after seeing an AI result, weaken its strongest feasible case, or treat an unresolved Report 1 cell as proof of benefit.

5.4 What Report 2 must be free to find

For any transferred function, Report 2 may find:

  • no AI intervention is necessary because completing P1 is sufficient;
  • a bounded AI intervention adds capability over P1;
  • claimed additionality is only adoption, speed or convenience and does not meet the outcome test;
  • AI creates ecological, democratic, distributional, agency, security or dependency failure;
  • evidence is insufficient and the result remains UNRESOLVED.

Those are genuinely open outcomes. Report 1 has not ranked P2–P4, assessed final delivery feasibility or made a trilogy verdict.

What this means

The answer is disciplined rather than dramatic. Australia has a comparator gap, not a demonstrated need for AI. Report 2 receives concrete functions, strong conventional alternatives, evidence gaps and failure rules. If an AI proposal cannot beat that comparator and pass every gate, the proper answer will be to strengthen P1 or leave the question unresolved.

Strongest Rival and Symmetric Comparison

The strongest rival says Australia can manage the portfolio through familiar public capabilities: good government, sufficient people, maintained infrastructure, enforceable law, mutual aid, trusted communication, social protection and democratic practice. It argues that many observed failures are implementation failures caused by fragmented authority, workforce scarcity, short funding cycles, weak monitoring and political choice—not shortages of computation.

That rival has strong evidence behind it. Australia already produces national risk and population assessments, coordinates emergency response, runs cyber exercises, funds recovery, reforms care standards, audits public agencies and conducts elections with established integrity institutions. The aged-care workforce measure improved in one year. IP Australia shows that narrow technology use can operate inside maturing governance. The OECD trust data show a democratic foundation that is neither negligible nor uniformly weak.

The rival’s weakness is not imagination but demonstrated delivery across interacting conditions. It must explain why repeated recommendations will now close, how prevention will outrun exposure, how regional care will be staffed, how essential supply will be tested, how democratic influence will be more than consultation, and how ecological decline will reverse. The dated backcast gives it a fair opportunity to do so.

Symmetry requires the same standard later for AI. P1 does not get credit for a plan, and P2–P4 cannot get credit for a model. P1 is charged for workforce, energy, procurement, delay and political feasibility; AI pathways must also be charged for data, compute, energy, vendor dependency, assurance, human oversight, security, opportunity cost and institutional learning. P1 benefits count where measured; AI benefits must also be measured. Harms and counterevidence are recorded for both.

Under that symmetry, the present conclusion remains stable: P1 is not proven adequate, but the strongest rival is sufficiently credible that additional AI cannot be presumed necessary.

Evidence Limits and Counterevidence

The evidence base has five important limits.

First, national reporting is uneven. Climate, population and some service data are strong, while protected national-security detail and cross-sector recovery results are necessarily less public. A public-interest report must not mistake secrecy for absence, but it also cannot certify what it cannot verify. The full 2025 Critical Infrastructure Annual Risk Review and exact S1 civilian recovery evidence remain unresolved.

Second, time periods differ. The biodiversity assessment is from 2021, disability and some unmet-need data from 2022, the Intergenerational Report from 2023, and other evidence from 2024–26. Later reforms may already be improving results. The backcast is designed to capture that rather than freeze Australia in its weakest year.

Third, attribution is hard in interacting systems. An avoided disaster, maintained service or trusted decision often has several causes. This report therefore uses contribution and continuity evidence rather than pretending every outcome has a single policy cause. It also distinguishes national transfer from agency examples: the ATO and IP Australia audits reveal governance patterns, but neither represents all government or industry.

Fourth, group data do not cover every intersection. Survey exclusions, small samples and inconsistent administrative categories limit remote, institutional and intersectional analysis. The scorecard does not fill those gaps with national averages; it assigns owners and checkpoints.

Fifth, the strongest feasible comparator is a structured counterfactual, not an observed national package. Its feasibility is grounded in existing authorities and programs, but its combined 2035 outcome cannot be known in 2026. That is why the verdict is “not demonstrated”, not a claim that adequacy is impossible.

Counterevidence is material. Australia’s response institutions work at scale. Disaster-funding assurance was largely effective even when performance monitoring was weak. The working-age population is projected to keep growing. Residential care-minute compliance improved. Commonwealth cyber plan coverage is high. IP Australia’s AI use was largely effective. Targeted ecological measures help some species. These facts prevent a simplistic decline narrative and explain why many cells are “partly meets” rather than “does not meet”.

Where This Argument Could Be Wrong

The conclusion would be too pessimistic if protected or newly published evidence shows Australia has already tested and met S1 civilian continuity and multi-risk recovery thresholds; if 2025–26 reforms rapidly convert climate plans, care standards, automated-decision rules and platform governance into measured outcomes; or if the ecological baseline has materially improved since the 2021 assessment. Those changes should move the relevant cells rather than be resisted to protect the thesis.

The conclusion could be too optimistic if current reporting understates compounding infrastructure failure, ecological thresholds, workforce exit, unpaid-carer burden, vendor concentration or information manipulation; if apparent response capacity depends on drawing the same people and assets into simultaneous crises; or if future fiscal and political constraints make the strongest comparator infeasible.

The method could also be wrong. No averaging is intentionally conservative. A decision-maker might accept trade-offs the method forbids, or reject a veto as too absolute. Conversely, the selected thresholds may still be too permissive where irreversible ecological loss, rights or sovereignty are involved. The series makes those value choices explicit so they can be challenged rather than hidden in a composite score.

Finally, the interaction screen may miss mechanisms. The six pairwise relationships do not fully model three- or four-risk feedbacks. Pairwise support is used as an eligibility gate, not a complete systems model. The 2033 multi-risk checkpoint is the practical test for higher-order coupling.

Implications — Separate from the Verdict

The verdict itself is only that P1 is not demonstrated adequate and has comparator gaps. The following are implications, not additional findings.

For governments, the immediate implication is to improve P1 regardless of later AI choices: publish outcome baselines; turn prevention and care commitments into capability; test essential-service recovery; give learning actions deadlines; and create representative response duties. These are no-regrets institutional tasks within the comparator.

For Report 2, the implication is a high burden of proof. It should begin with the specific gap, state why further P1 reform is insufficient, describe the AI intervention narrowly, and test material additionality. It must include already embedded AI in the baseline and avoid counting adoption or faster processing as living-well benefit.

For public debate, the implication is to resist two shortcuts: “Australia already has systems, so it will cope” and “systems have gaps, so AI is required”. Both skip implementation, distribution and counterfactual evidence.

For evidence owners, the implication is that transparent assurance can improve democratic legitimacy without revealing protected details. A publishable finding that a threshold was independently tested and met is more valuable than either complete secrecy or operational disclosure.

The Bottom Line

Australia possesses many of the institutions needed to manage the interacting risks to 2035. Its strongest feasible non-AI reform package is substantial and deserves full credit. Yet the evidence does not show that this package will meet every critical function, condition, domain and distribution requirement. Ecological decline, care and workforce gaps, implementation weaknesses, information-governance limits and unresolved continuity tests prevent an overall pass.

The provisional Report 1 answer is therefore: comparator gap, not AI necessity. Report 2 may test whether AI adds anything necessary and safe. It inherits no presumption that it does.

Acknowledgements

This report relies on public institutions, auditors, statisticians, scientists, researchers, service providers and communities whose records make independent scrutiny possible. Acknowledgement and source inclusion do not imply endorsement by any cited institution.

Authors

Rick Molony, Our Resilient World. AI assisted the research and drafting under human Product Owner direction.

Appendix A — References

Source IDs resolve to the evidence ledger. Access and verification date: 22 August 2026 unless stated otherwise.

Appendix B — Glossary

Adequacy — Delivery and continuity of a required function across every mandatory domain, condition and material group, with vetoes and gates passed.

Baseline — Australia’s observed 2000–2025 record and the systems operating at launch in 2026, including already embedded AI.

Comparator gap — A P1 function that does not meet, partly meets, is uncertain or is unresolved under the accepted test. It is a question for later comparison, not proof that AI is needed.

Continuity — The ability to maintain or restore a function under the relevant condition, including accessible fallback and fair recovery.

Embedded AI — AI already adopted in markets, government or infrastructure by the 2026 launch date. It remains part of P1 rather than being imagined away.

No averaging — The rule that a strong result in one domain, condition or group cannot cancel a mandatory failure elsewhere.

No exported harm — The rule that burdens shifted to another group, place, generation, worker, carer, ecosystem or supply chain still count.

Outcome — A measured change in capability, continuity or lived/system condition; it is not synonymous with announcement, spending, activity or adoption.

Portfolio-critical — A function made decision-critical by at least one supported or qualified risk interaction and therefore subject to the full adequacy test.

Strongest feasible non-AI comparator — The good-faith package of institutional, workforce, public-service, regulatory, material, democratic and conventional-technology reforms that Australia could plausibly deliver by 2035.

Veto — A mandatory ecological or democratic/information failure that blocks an overall adequacy pass and cannot be averaged away.

UNRESOLVED — Comparator-grade evidence was not located or cannot support a judgement; an owner and checkpoint are required.