An individual's AI opportunity

Executive summary

Develop a personal, recurring AI service allowance that an eligible person can use immediately for self-chosen lawful activity, independently of employers and educational institutions. Pay participating suppliers on the person's instruction, provide optional help, and make business backing a separate voluntary route. No learning test, course, business registration or political participation should determine basic access.

The strongest overseas precedents supply different pieces of this design. China offers local OpenClaw support proposals and a newer multi-model subsidy platform; Singapore offers course-linked premium AI access; India offers a compute marketplace with eligibility and project approval; US and UK accounts separate an individual's entitlement from service administration; and philanthropic programmes back self-directed projects. None of the examples reviewed demonstrates the entire proposed national entitlement or its Australian employment and cohesion effects. [B01–B13]

The new finding since the brief is Beijing E-Town's July 2026 real-time model-voucher platform. It makes an immediate-settlement mechanism more concrete, but eligibility is for enterprises. Another useful addition is the 1517 Medici Project: small project grants without an equity/IP claim, followed by possible later investment. These are design precedents, not reasons to turn individuals into businesses before helping them. [B03, B06]

Recommend age 16+ with no upper limit as the working case and A$1,200 a year as the central service-allowance scenario to price in B2, alongside A$600 and A$2,400. These remain recommendations for review, not approved expenditure or a claim that any amount purchases unlimited compute. Support, connectivity and accessibility require separately visible costs.

Current judgement: DESIGN PLAUSIBLE; DELIVERY UNRESOLVED. The individual journey is concrete enough for detailed institutional and fiscal design. It is not yet ready to open accounts, award funds or promise national rollout.

Orientation and accepted premises

B1 is the first of three individual-opportunity reports. It specifies what a person should receive. B2 will compare administrators, suppliers, infrastructure, funding and investment terms. B3 will integrate that design with working-time arrangements and a proposed One World Community demonstration through 2035.

P1–P3 remain ASSUMED FOR DESIGN: some people use shorter paid hours to experiment when AI is free; personal learning by doing is the programme's foundation; and capability contributes to resilience against job disruption and malicious-AI cohesion harm. We investigate implementation, reach, costs and consequences without requiring a preliminary proof of those premises. Adverse evidence still matters to the design.

Citizens and permanent residents, individual beneficiaries, no new government organisation to manage funds, and AI assessment with human final investment decisions are settled. Minimum age, overseas-residence treatment, allowance level, administration and investment terms are not silently treated as approved.

Confidence is high in verified programme rules and announcements as descriptions of those programmes; medium that their mechanisms can inform an Australian design; and unresolved for this proposal's scale, fiscal sustainability and national effects. Programme-owner publicity is not independent impact evidence.

Pass 0 — Define personal opportunity

The entitlement belongs to the person across school, work, unemployment, caring, self-employment and retirement. Paying a supplier does not make that supplier the intended beneficiary. Institutions may supply services; they must not decide which citizens deserve the initial opportunity based on a career plan or preferred learning topic.

Outside employment means a personally controlled account, choices and learning activity independent of employer direction. It does not require tracking a person's physical location or forbidding voluntary use of skills at work. Employers should have no access to private prompts or authority to withdraw the entitlement. The programme should not depend on unpaid work for an employer masquerading as personal learning.

Free at the point of use means no personal payment within the published funded allowance. It is incompatible with reimbursement that first requires a credit card, and with an agent silently running up personal charges. Unconditional initial opportunity means no performance or behavioural conditions beyond the stated eligibility, lawful-use boundary and necessary account/payment operation. It does not mean an infinite public budget or the absence of privacy and consumer protections.

National AI infrastructure means the capacity, interfaces, contracts and continuity arrangements that make the opportunity usable. A GPU allocation requiring specialist deployment is a different product from an accessible personal AI service. A person should not need to understand tokens, API keys or model hosting before beginning.

Pass 1 — Overseas mechanisms and their transfer limits

China: distinguish drafts, operating mechanisms and outcomes

Longgang's government-hosted March account describes a draft offering free deployment, three months' compute for solo startups, a 30% model-usage subsidy capped at RMB1 million annually, and potential equity backing for selected seed projects. Wuxi High-Tech Zone's issuing bureau separately published a March consultation on OpenClaw/OPC measures. No final instrument or outcome evaluation for those two particular packages was verified here. They are local, selective mechanisms, not an entitlement for every Chinese resident. [B01–B02]

Beijing E-Town's official July account states that a multi-model voucher platform launched on 14 July 2026. Enterprises legally operating in the area can obtain subsidies of up to 65%, with unified billing and immediate settlement rather than waiting for reimbursement. This is stronger evidence of an operating access mechanism than a draft consultation. The account does not independently establish actual take-up, fiscal performance or employment effects. [B03]

The Australian adaptation is a personal right to direct settlement across competing services. Do not import company registration, relocation or a commercial-project test. OpenClaw is an example of an agent framework, not a required provider. Simple chat, creative tools, coding, open models and agent experimentation should be possible within an explicitly priced service offer. Subsidised installation alone is insufficient if inference and hosting remain unaffordable.

AI access in Singapore and India

Singapore's government describes six months of premium AI access for citizens enrolling in selected SkillsFuture AI courses, scheduled from the second half of 2026. This connects learning with real use, but retains a training gateway and time-limited access; the announcement alone is not a verified usage or persistence result. Borrow funded access and sustained-use measurement, not compulsory enrolment. [B04]

IndiaAI's February 2025 end-user policy describes a shared marketplace and categories including researchers, students and early startups. Sections 3–4 require eligibility evidence and project/approval processes. “Independent researcher” does not mean any citizen with no qualification. Borrow supplier aggregation and transparent resource purchasing; do not import these filters into base personal access. GPU capacity and consumer service usability must remain separate requirements. [B05]

Funding outside traditional educational hierarchies

These examples support a design combining broad opportunity with a separate path for ideas requiring more resources. They do not show that selected exceptional founders represent everyone, that all people want to build a company, or that public returns will finance the entitlement.

What the US child and education mechanisms contribute

The UK Child Trust Fund is another useful age/account precedent: beneficiaries can control an account at 16 and withdraw at 18; the scheme closed to new entrants in 2011. It separates ownership, control and access, without demonstrating the right age for AI funding. [B14]

Pass 2 — Specify the minimum useful personal service

The proposed offer has seven requirements:

1. Personal activation. Confirm citizenship/permanent residence and age through an accessible process. Offer a practical alternative for people who lack a smartphone or conventional digital identity. No essay about ambition or proof of disadvantage.
2. Immediate use. An eligible person chooses a service and starts without paying first. The wallet displays remaining allowance in Australian dollars; advanced metering is available for those who want it.
3. Real choice. Access to more than one participating provider, with a usable open-model option considered in procurement. Do not promise every proprietary model will participate.
4. Self-chosen activity. Creative, practical, social, technical and non-commercial uses are legitimate. A person does not need to label an activity as education, employment preparation or business development.
5. Controlled expenditure. Clear limits and user-set budgets for experiments; no automatic personal debt or silent paid renewal. On exhaustion, retain access to the person's exported work and explain continuation options.
6. Optional assistance. Accessible setup help, peers, examples and learning supports available without a compulsory course or a test. Help can come from community organisations, libraries or existing providers without making their membership an eligibility gate.
7. Portability and privacy. The person can change suppliers and export their own material where rights and formats permit. Payment administration should not collect private prompt histories to rank deservingness.

These are procurement and service requirements proposed for B2. They are not statements that an existing Australian platform already satisfies them. In particular, model availability, account portability, privacy terms and access for minors need provider-by-provider verification before a funded offer can be promised.

Age recommendation and reach

Start at 16, with no upper age limit, as the central design case. It preserves the original “over fifteen” intent and reaches people before many workforce-entry decisions. Starting at 15 offers earlier access but enlarges the under-18 implementation task; starting at 18 simplifies some contracting but leaves a gap at earlier transitions. The international precedents show that age arrangements can be separated; they do not identify an optimal cutoff.

For 16–17-year-olds, specify a lawful age-appropriate route with any necessary consent/contract arrangements. Include a route for young people without supportive adults. Do not present school approval or parental wealth as an eligibility test. Business funding contracts may need a different supported route; initial learning access must not be held back while those contracts are designed. This is a design recommendation requiring Australian/provider-specific legal work, not a claim that minors can already enter every contract independently.

Recommend using Australian-resident citizens and permanent residents as B2's principal delivery/costing case and separately costing overseas citizens. Residence-abroad treatment remains a human design choice: the recommendation narrows the administrative working case, not the citizenship decision already settled.

Three illustrative journeys

Young first entrant: a 17-year-old chooses to build a game or understand an unfamiliar occupation. They can try, compare outputs, make mistakes and seek help without needing a business plan. They control their project within the applicable minor-account arrangements.

Older worker: a 59-year-old experiments with a personal maintenance guide, investigates a new role or explores a microbusiness. The programme does not require leaving employment, accepting self-employment or showing that AI will replace the current job.

Carer outside paid employment: the person uses voice or other accessible tools in short sessions and may prefer a creative activity. A four-day-week reform does not release paid hours for them, but the entitlement remains the same. If equipment, data or accessibility prevents usable access, a nominal compute credit is not a successful service.

These are design scenarios, not participant results. They reveal requirements that a business-only or course-only scheme would miss.

Pass 3 — Personal learning, malicious AI and cohesion

The 2025 Australian Digital Inclusion Index AI case study reports a marked age difference in recent use: 69.1% among 18–34-year-olds versus 15.5% among 65–74-year-olds. Use is not competence, but the finding makes a digitally self-service-only rollout a poor assumption. Offer accessible help and observe who cannot convert nominal access into usable opportunity. [B15]

Learning by doing can involve testing an answer, trying an alternative, asking for an explanation and checking whether the result works. An optional “help me learn” mode could scaffold those actions without preventing people from simply using a tool. A person should retain control over how much explanation they want.

Relevant counterevidence is retained. Bastani and colleagues' school-mathematics experiment found that unrestricted-style AI assistance could impair later unaided performance; a hint-oriented tutor mitigated the problem. The August 2025 correction concerns an author affiliation, not the reported results. Kestin and colleagues' 194-student physics trial found benefits from a deliberately designed AI tutor compared with active learning. Neither is an Australian personal-career programme. They support distinguishing assisted output from learning and keeping optional instructional help; they do not justify making formal education a gateway or declaring all education useless. [B16–B17]

For malicious-AI resilience, offer personally chosen practice in checking sources, verifying an unexpected request through a separate known channel, recognising uncertainty and recovering from mistakes. Use fictional or consented material for demonstrations. Do not teach that a model can certify truth or that a person can reliably identify a deepfake by appearance. eSafety's account of deepfakes describes abuse and deception risks; it does not establish that increased AI use alone prevents them. [B18]

Measure information judgement, overconfidence, recovery, belonging, relationships and perceived fairness separately from prompt counts and political agreement. A person who disagrees with the programme must remain eligible. Optional peer learning can encourage human contact, but it must not become ideological selection or a condition of financial support.

Provider safety and privacy duties should be implemented in service design, with understandable reasons and remedies. They must not become a discretionary government approval process for lawful personal interests. Some lawful activities may not be offered by particular commercial services; B2 must make those coverage limits visible and examine alternatives. No single supplier's terms should quietly be represented as the whole programme's definition of permissible learning.

Pass 4 — Funding an individual and then an idea

Base access and staged expansion

Keep two separate funding relationships. Base access is personal opportunity. It does not depend on measured progress and is not revoked when an idea fails. Expansion is optional backing for an idea requiring more resources. Assessment and accountable human judgement apply at that stage.

Recommend investigating three expansion steps in B2: a small validation grant; a larger staged project grant once evidence is produced; and investment where a viable business and appropriate terms justify it. This uses a venture-capital approach to uncertainty and staged commitment without forcing everyone into equity or debt. A business may eventually need a legal vehicle, but the base programme does not fund organisations as a competing class of beneficiary.

AI should prepare the substantive assessment: the intended user/problem, evidence of demand, feasibility, costs, competing explanations, risks, unknowns and the next experiment. The applicant can correct factual errors. Named people make every final investment decision, recording their reasons and any override. Multiple models can challenge each other, but correlated answers do not establish truth.

Research on Nigeria's YouWiN! competition found prediction of later business outcomes difficult for both judges and machine-learning approaches. That older, selected setting is not a test of contemporary generative models in Australia. It supports evaluating assessment quality prospectively rather than assuming an articulate model-written business plan predicts success. Do not import demographic predictors as eligibility rules. [B19]

A worked assessment example — simulated only

An individual proposes an AI-assisted inventory tool for small repair shops. The only evidence is two informal conversations and a mock-up. A model assessment should identify the absence of paying-customer evidence, uncertain support costs, data permissions and alternatives already available. It can propose a small test with consenting shops and a capped resource request. It must not invent sales or declare the applicant investable because several models agree.

The human decision-maker could authorise a bounded validation grant, reject it or request changes. This report makes no such real award. Later stages would examine actual demand and net income, with failure tolerated within the agreed budget. The person retains their general AI access whichever decision is made.

Account and infrastructure responsibility

Australia already has a National AI Plan, launched in December 2025, with opportunity, benefit and safety goals. Its existence is context for B2, not evidence that this proposed entitlement is already funded. [B20] myGov provides access to linked government services; that does not make it an existing AI-wallet payment system. [B21]

B2 should compare an existing public institution with contracted settlement, a competitively contracted independent administrator, and a non-government trust/cooperative route with an explicit public-funding contract. Each must answer who verifies entitlement, pays suppliers, handles disputes, reports expenditure and takes responsibility for investment. No new government organisation is proposed. A label such as “platform” cannot conceal a new approval bureaucracy.

National infrastructure options include purchased multi-provider inference, accessible open-model hosting, reserved domestic capacity and diversified external supply. Assess resilience to supplier exit, model changes, price increases, outages and hardware dependencies. Domestic hosting may improve control over some risks while retaining other foreign dependencies. Add energy, water and local public-benefit requirements from the prior ORW infrastructure research. No new frontier-model training project is presumed necessary for personal access.

Pass 5 — Cost sensitivity and the route to 2035

Let N be eligible people, u the active-use share, C the annual individual service ceiling and r the average redeemed fraction among active users. Expected base-service cash cost is N × u × r × C. Maximum annual base entitlement exposure is N × C where everyone can redeem the ceiling. Add support, administration, infrastructure and evaluation separately; add venture commitments with their disbursement timing. Do not deduct speculative investment returns or avoided job losses.

The table assumes r = 1 and normalises to one million eligible people. It is not an estimate of Australia's eligible population.

Recommend pricing the A$1,200 central case against a concrete basket of ordinary chat/research, creative work, coding and bounded agent experiments. The lower and higher cases remain live alternatives. These are policy scenarios, not current vendor quotations or a judgement that the middle amount is sufficient. Heavy compute can exhaust even the higher allowance quickly. B2 must obtain reproducible service-price assumptions, use realistic eligible-population data, model partial redemption and reserve capacity for higher uptake. Inflation and future prices are unknown; 2026-dollar scenarios are not a nominal 2035 appropriation.

A recurring annual allowance is more consistent with a continuing 2035 opportunity than a short promotional credit followed by an abrupt payment cliff. Recommend user-controlled use across the year and analyse bounded rollover in B2; do not decide the rollover rule silently. Separate accessibility/support budgets avoid giving a person less experimentation simply because their setup takes more work. Device/connectivity assistance is a proposed enabling option to cost, not an approved expansion of the compute allowance.

Four pathways — the individual's opportunity

The common WF1–WF4 definitions apply to B1 as well as A1. The following compares designed experiences; overseas programme accounts do not establish these as observed Australian outcomes.

Each pathway needs the same cost boundary: existing and incremental service expenditure, time/coverage costs, support, infrastructure, administration, evaluation and separately timed project commitments. The A$600/A$1,200/A$2,400 rows are allowance sensitivities, not WF2/WF3/WF4 labels. Choosing the largest allowance does not define success. B2 must cost each complete package against the same population and prices, avoiding double counting shared infrastructure or assumed returns.

Build WF4 through specific changes: unusable nominal access → accessible activation and funded optional support → accountable administrator and providers → B2 service/budget specification and later observed activation → redesign if exclusion persists; supplier dependence → diversified capacity and tested switching → contracted operators → continuity evidence before expansion → preserve personal data/export access and avoid personal debt during failure; assisted output without transferable capability → optional verification/learning supports → participant-controlled services and independent evaluation → B3 novel-task and resilience outcomes → revise delivery if outcomes fail to improve. Time and service barriers are jointly owned by A2. No new government funding organisation is created by assigning these responsibilities.

Apply price shocks, fiscal constraints, faster displacement and deception risks to all pathways. WF4 must demonstrate a workable response to those conditions; it cannot assume cheap compute, unlimited public funds or that malicious actors stand still. If access is delivered but the defined employment or cohesion outcomes remain unsupported, report partial delivery and the evidence gap rather than declaring full success.

Backcast and demonstration route

This is the WF4 individual-opportunity backcast. WF3 can deliver the personal service while time reform is incomplete; that useful route must be compared with WF1 and WF2 rather than conflated with the integrated endpoint. The common programme success criteria govern the 2035 assessment.

Within a later authorised demonstration, use consenting participants to assess activation burden, novel-task performance, verification, transfer, expenses, actual hours, net income and cohesion outcomes. Include people who decline evaluation in ordinary entitlement access. A comparative time-only/access-only/combined design can investigate practical effects without making proof of P1–P3 the entry gate for developing this proposal. Randomised or phased research offers require their own justified protocol; they must not quietly remove an already established entitlement.

Accountability is proposed, not appointed: the human research owner owns the design judgement; the research team and One World Community develop and review the priced proposal; an existing authorised funder owns any later appropriation or funding contract; a selected administrator owns settlement and continuity; providers own contracted service delivery; named human panels own project awards; independent evaluators own outcome reports. Readiness requires the relevant decision record, priced service, funding instrument, operational agreement and measured performance, rather than a model-generated assurance.

If working-time reform is delayed, personal access can proceed under its own authorised funding route. If national funding is delayed, a separately resourced limited demonstration could examine service operation but must not be advertised as a national entitlement. If supplier prices rise sharply or a provider exits, use the continuity plan, publish the revised purchasing power and reprice the offer before expanding. Provider failure must not create personal debt. These are alternative implementation paths, not promises that funding exists.

Objections, trade-offs and handoff

“Free consumer tools already solve this.” They are a serious baseline to compare at equal public cost. B2 must identify the additional usable capability, continuity and choice purchased by the entitlement. A subsidy that buys no additional opportunity should be redesigned rather than defended because it is labelled national infrastructure.

“Funding everyone wastes money on people who already pay.” Some substitution for private spending is likely in principle, but its magnitude is unknown. Universal eligibility favours simple access; targeting may reduce expenditure while introducing conditions excluded by the stated design premises. Show that trade-off and fiscal exposure openly. Activation can be voluntary without a merit test.

“Personal freedom conflicts with an auditable public programme.” Audit identity, settlement, pricing and programme outcomes with appropriate data, rather than demanding private learning diaries. Legal and fraud controls must be designed proportionately. Do not represent a surveillance-heavy solution as the unavoidable consequence of accountability.

“AI assessment eliminates the need for judgement.” It can structure and challenge evidence; it cannot establish absent customers, bear public responsibility or remove uncertainty. Final human decisions and applicant correction are integral to the requested approach.

“The four-day week must come first.” People can receive personal access before national hours reform. A1 shows why time arrangements need sector-specific work. The shared 2035 programme should not make non-employed people or slower-transition sectors wait for the largest institutional reform.

B2 handoff: cost the recommended age/service scenarios, confirm residence treatment, compare actual administrators without a new government body, price usable multi-provider access, analyse accessibility and rollover, and specify staged grants/investments with AI assessment and human authority. B3 handoff: design a voluntary demonstration and national outcome/continuity framework using A2's time assumptions. Neither receives a proven employment effect, a funded budget or an approved supplier from this report.

What this design can justify now

Detailed funding and institutional comparison depends on human review of the proposed personal offer. The unresolved choices are age/residence, the costing envelope, and separating unrestricted initial opportunity from assessed project expansion. The individual-learning and human-decision principles remain settled design premises.

This is not a worldwide ranking of “best practice”. It is a mechanism review spanning national and local government, education-related accounts and independent project backing. Unverified disbursement, outcome and transfer claims remain visible. No national unconditional multi-provider entitlement matching this design was verified among the selected examples; that is a search result, not proof that none exists anywhere.

References and source locations

Retrieved 9 September 2026; programme dates and evidence status are distinguished in the ledger. Operator accounts establish stated mechanisms, not independently evaluated impact.

• B01: Longgang government-hosted account, 11 March 2026, draft and solo-startup paragraphs.
• B02: Wuxi High-Tech Zone issuing-bureau consultation, 18 March 2026, notice body; Chinese text paraphrased with AI assistance; not an official translation.
• B03: Beijing E-Town real-time model-voucher platform, 20 July 2026 account, launch, enterprise eligibility and settlement paragraphs.
• B04: Singapore government: support through the AI transition, selected SkillsFuture courses and six-month access.
• B05: IndiaAI End-User Policy for AI Services on the Cloud, February 2025, §§3–4 and payment terms; current portal eligibility.
• B06: 1517 Medici Project, stipend, ownership, non-startup and geographic boundaries.
• B07: Mercatus: Emergent Ventures, eligibility and programme description.
• B08: Thiel Fellowship, current grant and college stop-out terms.
• B09: IRS Form 4547 instructions, account versus seed eligibility, growth period and distributions.
• B10: IRS Topic 313: qualified tuition programmes, beneficiary contributions and qualified uses.
• B11: Arizona ESA support, ClassWallet and spending administration.
• B12: IRS Federal Scholarship Tax Credit, 2027 commencement and participating-state mechanism; current official indexed text retrieved.
• B13: Kela: evaluation of Finland's basic-income experiment, population, evaluation and outcome summary; not a compute experiment.
• B14: UK Child Trust Fund overview, control/withdrawal ages and scheme closure.
• B15: Australian Digital Inclusion Index: the AI divide, age comparison; self-reported recent use.
• B16: Bastani et al., PNAS (2025); affiliation correction. Full-text XML retrieved through Europe PMC; summary and correction checked, no independent statistical replication.
• B17: Kestin et al., Scientific Reports (2025). Full-text XML retrieved through Europe PMC; abstract/design and population checked.
• B18: eSafety: Deepfakes, risk and response account; evidence of a threat, not programme efficacy.
• B19: McKenzie and Sansone, World Bank Working Paper 8271 (2017), abstract, introduction and study context; later published in 2019 as Predicting entrepreneurial success is hard. Historical machine learning is not current generative AI.
• B20: Australian National AI Plan, launch date and goals; detailed programme fit is B2 work.
• B21: myGov: create an account, existing account function; no AI settlement capability inferred.

Glossary

Entitlement: an eligible person's defined opportunity to access support. Settlement: payment to a supplier for the individual's authorised purchase. Inference: running a trained model to produce outputs. Compute allowance: a funded usage/service ceiling, not a promise of unlimited resources. OPC: one-person company. Additionality: benefit beyond what would otherwise occur. Redemption: use of available funding. Backcast: a dependency-based implementation route from the desired future.

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.