Funding people and their ideas

Executive summary

Design a personal service entitlement funded through an existing Commonwealth portfolio, with contracted payment administration and competing AI suppliers. The individual chooses lawful uses; a course, employer, company or assessment of ambition does not determine basic access. Existing public institutions remain accountable for public funds. No new government organisation is proposed.

Keep age 16+ as the working case, but procure a verified service for 16–17-year-olds. Current product terms make a universal adult-subscription voucher inadequate. Anthropic describes an API route for organisations serving minors with additional requirements; Google's Gemini Developer API terms restrict both under-18 access and the consumer-use case. These are material procurement constraints, not reasons to require school enrolment.

An illustrative annual scenario with 20 million eligible people, 50% active use, 75% redemption and a A$1,200 ceiling costs A$9 billion for service use. Adding the declared administration, help, infrastructure, device, evaluation and project-backing assumptions gives A$10.6 billion annually, plus A$100 million assumed setup. At full participation/redemption the same model gives A$26.55 billion annually. These are explicit budget scenarios, not approved expenditure or a verified eligible-population estimate. A2's working-time costs are additional.

Recommend ordinary, explicitly authorised public funding for the base offer; keep optional project grants and investment in a separately capped allocation, without forecasting venture returns as the means of paying everyone's bills. Retain A$600 and A$2,400 alternatives and compare complete service quality, not token volume alone. Develop WF4 as an integrated endpoint, with WF3 able to deliver personal opportunity while time reform advances.

Current judgement: DESIGN PLAUSIBLE; DELIVERY UNRESOLVED. The unresolved choices are the funding/administration direction, the scale of fiscal exposure and the separate under-18 route. The reports do not establish a ready operator, appropriation, national four-day-week cost or demonstrated employment/cohesion effect.

Reader orientation, premises and confidence

B1 specified the individual's experience and overseas mechanisms. B2 develops delivery and financial alternatives; B3 will integrate the demonstration and national outcome framework. A2 owns time/service costs. P1–P3 remain ASSUMED FOR DESIGN: voluntary experimentation with more time and free AI, personal learning by doing as the foundation, and a resilience contribution. They do not supply an uptake rate or a return on public investment.

Citizens/permanent residents, individual beneficiaries, self-chosen lawful use, no new government funding organisation and AI assessment with human final investment decisions are retained. Ages, amounts, residence details and contract terms are working proposals. Confidence is high in the sourced age arithmetic and stated current rules/prices, medium in the mechanism comparison, and unresolved in national fiscal adequacy and outcomes. Modelling inputs have no empirical confidence score where they are simply policy scenarios.

Pass 0 — Define eligibility, opportunity and the baseline

The proposed benefit is a recurring, person-controlled ceiling for eligible AI services. Suppliers receive settlement for the individual's authorised consumption; this does not make businesses a competing class of beneficiary. Individuals may use the service creatively, practically or commercially without proving job risk or educational progress.

For costing, use Australia-resident citizens and permanent residents as the central population case. Overseas eligibility remains a separate decision. Eligibility needs an operational definition and lawful verification; living indefinitely in Australia is not automatically equivalent to holding a permanent visa. Home Affairs describes the New Zealand Special Category Visa as temporary despite its work/stay rights, with separate citizenship arrangements. Do not silently include or exclude that group by using an ambiguous population proxy. [B2-15]

What the population data establish

The ABS December 2025 release, published June 2026, reports 27,801,023 residents at December 2025. Its annual single-age workbook supplies revised June 2025 counts. Summing the Persons series gives the following resident, not citizen/PR, denominators. The workbook and extraction locations are recorded in the model inputs and age CSV. [B2-01]

No matched current citizenship/permanent-residence-by-age count was established. Use 18 million and 20 million as explicit eligibility sensitivities, with 22,467,512 as the reference-date resident envelope. It is not a 2035 ceiling or a validated eligibility forecast. Do not subtract a different-date temporary-visa stock from ERP and call the result exact: coverage and residence definitions differ. Refresh matched data before a national appropriation and model population change separately from service-price change.

Dated WF1 inventory

As at 9 September, relevant existing approaches include private/free AI services, employer-led adoption, education/training, government digital-service administration and business-focused AI assistance. The AI Adopt grant-recipient list identifies centres serving SMEs; it does not establish a universal personal compute entitlement. The National AI Plan and March 2026 infrastructure expectations are policy context, not a funded individual wallet. [B2-02, B2-08]

Retain existing spending as B0. The baseline inventory does not reconcile every Commonwealth/state education and AI budget, and this report assumes none is cancelled. The new programme's incremental cost is added to B0; the WF1 incremental value of zero never means government currently spends nothing. WF2's additional conventional training is separately costed as a scenario, not hidden in personal compute.

Pass 1 — A usable personal service and a separate under-18 route

Recommend a person-initiated annual account with a 365-day allowance, immediate supplier settlement and understandable remaining funds. Opening the account should not require a personal credit card or upfront reimbursement. The person can choose among participating subscriptions, metered services and an accessible open-model service where licensed and operationally suitable. No provider is selected by this report.

Prefer a full annual ceiling available on activation, with user-set experiment limits and a hard no-debt boundary. Compare this with monthly release, which manages demand but can frustrate a concentrated project. Recommend no monetary rollover in the initial costed specification; preserve the person's files and renewal opportunity. A 25% one-year carryover alternative would increase possible base exposure to 1.25 × N × C and needs separate funding. Unused allowance does not create an investment claim or evidence that the person is undeserving.

Optional setup/accessibility help is funded separately. Provide routes for people with limited digital access, without supportive adults, or unable to use a conventional smartphone identity flow. Libraries/community organisations can deliver support under a service contract, but membership and course attendance remain optional. Identity/payment audit should not collect private prompt histories to score deservingness.

Eligibility does not guarantee supplier availability

Anthropic's consumer terms require adults. Its current guidance describes additional measures for organisations offering API-based services to minors, including appropriate safeguards and AI disclosure. This supplies a candidate route for an independently controlled 16–17-year-old account through a contracted service; it is not confirmation that an existing operator meets Australian requirements. [B2-12, B2-13]

Gemini Developer API terms effective March 2026 require users to be 18+, exclude API clients directed at or likely to be accessed by minors, and specify professional/business rather than consumer use. Its API tariff can inform price comparisons for eligible use; it is not a universal personal-service offer. Do not route minors through a shared adult account or assume paid access overrides those terms. [B2-11]

Recommend retaining 16+ and specifying a compliant youth service rather than raising the age merely for procurement convenience. Adult consumer access, commercial APIs, hosted open models and negotiated services require separate verification. The basic lawful-use intent remains; supplier restrictions and coverage gaps must be visible. No one supplier should become the final definition of which lawful interests the programme permits.

Purchasing power at current listed prices

The following are illustrative token workloads, not equivalent-quality products, complete service quotes or recommended latest models. They use two listed tariffs: Gemini 2.5 Flash-Lite at US$0.10 input/US$0.40 output per million tokens, and Claude Sonnet 4.6 at US$3/US$15. Neither price is a 2035 commitment. [B2-09, B2-10]

These gross-charge scenarios assume US$1 = A$1.50 and a 1.10 tax factor; neither is asserted as the current exchange rate or the government's tax treatment. The CSV also varies exchange rates. Tool calls, search, storage, interface hosting, image/video/audio generation, retries, long-context effects and support can add substantial cost. Tokenisation differs across models, so a fixed token count is a billing example rather than an equal-work benchmark. Public pricing does not establish redistribution rights or procurement availability.

A$1,200 annually is A$100 monthly on average. The builder example shows how one active experiment can consume that monthly equivalent; low-cost text use can consume much less. This supports retaining meaningful choice and separate assessed expansion. It does not establish A$1,200 as the optimal allowance. Before delivery, price an end-to-end basket including creative and agent activity, measure quality and accessibility, and compare with free services. Do not require a person to use the benchmark basket.

Pass 2 — Administration without a new government organisation

Finance distinguishes legal authority, appropriations and the rules governing grants. Contracting a third party does not remove the Commonwealth's responsibilities. Procurement has its own value-for-money framework. A personal service entitlement, supplier procurement and discretionary venture awards need separate legal/financial treatment; simply calling all three a wallet does not settle that classification. [B2-03, B2-04, B2-07]

Services Australia states an existing service/payment delivery role; DTA's BuyICT connects government purchasers and ICT suppliers. These are relevant institutional functions, not evidence that either body has accepted this programme or that an existing panel automatically covers every proposed contract. A policy portfolio such as Industry is a candidate sponsor; the actual allocation of functions requires government decision and costed operating design. [B2-05, B2-06]

For the recommended route, distinguish roles: the existing funder owns entitlement and budget; the administrator verifies account facts and settles authorised purchases; providers deliver services; independent reviewers evaluate outcomes and complaints; named people make investment decisions. Individual access must not depend on a discretionary panel reviewing personal interests. A contracted company is paid for administration, not funded as the target beneficiary.

The One World Community can develop the public proposal, scrutinise delivery and provide optional support if resourced. It should not be presumed to possess national-scale financial infrastructure or required to become the gatekeeper to everyone's learning. Community involvement can be designed without a new government organisation or compulsory community membership.

Pass 3 — Infrastructure and accountable project backing

Buy dependable access before committing to a large build

Compare three infrastructure choices: diversified purchased services; a hybrid of purchased services and domestic hosted open models/reserved capacity; and extensive publicly funded dedicated infrastructure. Recommend the hybrid as the next design case, with capacity bought in stages against actual use. Ownership of a data centre is not the individual's benefit; dependable, accessible AI is.

B1's China/India examples show useful settlement and market mechanisms with eligibility limits. B2 retains that lesson: buy the function rather than importing a business-registration gateway. Domestic capacity should have explicit availability, continuity and public-benefit terms; imported hardware/model dependencies remain even when hosting is local. The March 2026 Australian infrastructure expectations identify national-benefit considerations; their existence is not an automatic enforceable guarantee in each project. [B2-08, B2-16]

Keep capacity availability charges separate from redeemed inference. If prepaid domestic capacity includes inference, subtract that portion from metered supplier payments; do not count it twice. Require priced exit/switching, data export, energy/water obligations, disaster continuity and a supported pathway when a model is withdrawn. The model's A$500m annual availability reserve is an explicit scenario, not a capacity quote or an approved construction project. High-cost dedicated frontier-model training is not assumed necessary.

Optional backing belongs to the individual and the idea

Use a staged portfolio with three proposed ceilings: A$5,000 validation grants, A$50,000 development grants, and investment up to A$200,000, with terms negotiated by authorised people. These are proposal amounts to review. Small grants should not automatically surrender an individual's IP or require equity. Larger investment can require a suitable legal vehicle and genuine rights/terms; basic AI access never depends on forming that vehicle.

An illustrative A$250m annual commitment pool could comprise 10,000 validation awards, 2,000 development awards and 500 investments at those ceilings. These counts are not forecasts and are not necessarily distinct people. Drawdowns should follow the approved project stages; accounting must distinguish commitments, cash disbursements and outstanding obligations. No return is deducted from the base entitlement's cost.

AI performs the substantive assessment: problem/customer evidence, feasibility, cost, alternatives, uncertainty, risks and the next useful test. It should challenge an applicant's claims and identify missing facts rather than reward a polished generated plan. The applicant can correct errors. People decide every final grant/investment award and any material change in terms; rejection must state reasons and offer an appropriate review route. Models do not acquire signing authority or silently reject applicants on inferred demographics.

A simulated repair-shop inventory proposal with a mock-up and two conversations might merit a bounded customer-validation experiment; it does not yet establish demand or investability. Models should expose that gap. Human decision-makers then consider the evidence, costs and available portfolio budget. A failed experiment can still have learning value and does not remove the person's underlying account. The earlier business-prediction evidence remains a reason to test assessment quality rather than assume AI scores predict success. [B2-16]

OAIC identifies automated-decision transparency obligations commencing 10 December 2026, including some computer-assisted decisions affecting rights or interests. A human final signature should not be assumed to remove the need for legal analysis of the AI assessment process. Plan understandable reasons, privacy and review from the outset; do not turn public accountability into routine disclosure of private learning prompts. [B2-14]

Pass 4 — Fiscal scenarios and the four pathways

Expected annual service expenditure is N × u × r × C: eligible population, active share, redeemed fraction among active people and annual ceiling. Base full-redemption exposure is N × C. Add actual administration, optional help, capacity, devices, evaluation and venture commitments. Support is not taken from the individual's compute allowance.

For the central model, administration is 5% of service spend; help reaches 10% of active people at A$200 each; annual allocations are A$500m infrastructure availability, A$100m devices/connectivity, A$100m evaluation and A$250m project commitments. Setup is A$100m once. Every one of these is a declared assumption requiring pricing and capacity validation. The resulting full-redemption package figure is a scenario, not a hard maximum if support needs or supplier costs change.

All exclude A2's time reform and retained BAU expenditure. The 90-row national scenario file varies population, uptake, redemption and allowance. With 18m eligible, the central A$1,200 package becomes A$9.635bn; using the June 2025 resident-age envelope gives about A$11.791bn under the same uptake/redemption assumptions. Those are alternative denominators, not an empirical confidence interval. Inflation and population/price changes to 2035 need separate scenarios; none of these values is a nominal 2035 appropriation.

Complete comparison, including what remains unpriced

The WF2 training allocation is a modelling choice, not a current programme budget. Its additional setup scenario is A$25m versus A$100m for WF3/WF4. Compare like populations and report distribution: WF2 may spend less partly because it reaches fewer people. Neither assumed uptake nor national full coverage is evidence of implementation.

Complete cost is retained baseline + incremental AI package + additional training where applicable + pathway time costs + capital/redesign not already included. Separate government outlays from employer/private costs and from transfers. A2 identifies how to estimate T2–T4; no reconciled national total exists here. A proposal quoting only A$10.6bn as the total cost of WF4 would be misleading.

Funding source and common stress cases

Recommend explicit general-revenue funding through an existing portfolio, with Parliament/government determining the authority and appropriation. Compare the opportunity cost with retaining existing spending; do not assume cancelling useful education finances the offer. A dedicated AI levy is a policy alternative requiring its own incidence, revenue and international analysis. No levy yield or promised budget saving is invented here.

Venture returns and philanthropy can supplement a separately funded programme but should not be booked as assured recurring income. As a purely arithmetic illustration, A$10.6bn annual funding at an assumed 4% endowment payout requires A$265bn capital, before uncertainty about investment returns and inflation. This is not a recommended investment return or endowment plan; it illustrates the scale mismatch with an ordinary demonstration fund.

Apply the same adverse conditions to all pathways: higher uptake, expensive models, exchange-rate depreciation, provider exit, constrained public finance, faster displacement and malicious use. At a fixed individual ceiling, price rises reduce purchasing power rather than mechanically increasing the base maximum; if policy promises a fixed service basket instead, expenditure can rise. Decide which promise is being funded. Contingency should preserve access to personal work and prevent debt, with explicit revisions to new commitments rather than hidden service degradation.

Pass 5 — WF4 backcast, review and handoff

B3 must pre-specify outcome comparisons and service acceptance criteria before any demonstration. Suggested operational requirements include activation without a personal payment, an accessible alternative channel, transparent remaining funds, no automatic debt, a verified youth offer and a tested supplier exit. Coverage, learning-transfer and cohesion effect thresholds require a justified protocol and review; there is no invented national pass percentage.

A3 handoff: include the complete time-plus-opportunity funding burden and understandable alternatives in the public proposal. B3 handoff: build the demonstration and outcome framework around the recommended individual service, optional project stages and common stress cases. Unresolved design choice: whether to develop the general-revenue/existing-portfolio route with contracted hybrid delivery and a separate youth service at this fiscal scale. No external engagement or award is inferred.

Objections and limitations

Does contracted delivery reintroduce organisational control? It can. The contract must leave the person's lawful choices independent of educational, employer or community selection, with payment administration limited to its authorised function.

Why fund people who can already afford AI? Universal eligibility trades some substitution for simple access and freedom from merit/income gates. Quantify additional opportunity and opportunity cost; do not assume all redemption is new learning. WF1/WF2 remain serious comparators.

Does the youth route alter the proposition? It changes service delivery to make the proposed age range workable; it does not require school attendance or parental wealth. Whether the service actually meets its requirements remains an implementation test.

Is this a complete procurement or budget submission? No. It is a source-backed design and reproducible fiscal envelope, with national time costs, matched eligibility data, operator quotes and legal instruments still to establish. The current judgement remains DESIGN PLAUSIBLE; DELIVERY UNRESOLVED.

References and source locations

Retrieved 9 September 2026. Prices and terms are dated snapshots, not commitments to 2035.

• B2-01: ABS, National, state and territory population, December 2025; Table 59 annual single-age workbook. Data1/Data2 Persons columns, latest annual row; June 2025 reference date.
• B2-02: AI Adopt grant recipients, centre beneficiaries and functions.
• B2-03: Finance, the grants process, authority and expenditure requirements.
• B2-04: Finance, Commonwealth grants FAQ, third-party administration obligations.
• B2-05: Services Australia, who we are, existing service/payment functions; no AI-wallet appointment inferred.
• B2-06: DTA, BuyICT, digital sourcing functions.
• B2-07: Finance, Commonwealth Procurement Rules, current framework; final route requires specific classification.
• B2-08: Industry, new data-centre expectations, March 2026, national-benefit context, read alongside B1's National AI Plan reference.
• B2-09: Claude Platform pricing, standard Sonnet 4.6 input/output rows; no caching/batch discount assumed.
• B2-10: Gemini Developer API pricing, standard Gemini 2.5 Flash-Lite text input/output rows.
• B2-11: Gemini API terms, effective March 2026 age and use restrictions.
• B2-12: Anthropic consumer terms, adult consumer eligibility.
• B2-13: Anthropic guidance for organisations serving minors, March 2026 guidance and service obligations.
• B2-14: OAIC, automated decision-making transparency consultation, December 2026 commencement; current official indexed text checked.
• B2-15: Home Affairs, Special Category Visa, visa classification; programme eligibility requires its own definition.
• B2-16: Programme B1, Report B1, overseas mechanisms and staged backing, including B03/B05/B06/B19. Prior evidence retained; no new programme-effect estimate.

Glossary

Appropriation: authority to draw public money for specified purposes. Settlement: paying a supplier for authorised usage. Inference: running a trained model. Redemption: consuming an available allowance. Commitment: an obligation to fund that may be paid over time. ERP: estimated resident population, distinct from programme eligibility. Additionality: benefit beyond what would otherwise happen.

Authorship and method

Research by Rick Molony, developed with AI assistance. This is bounded critical research and implementation analysis, not an independent audit, field trial, procurement quote, approved budget or legal opinion.

Suggested citation

How this page was derived: the full substantive source body was rendered after the declared public-context transformations, removal of private delivery/workflow material and omission of the private change log.