Summary of recommended actions

The United Kingdom and European governments should prepare now for the possibility that AI weakens wage labour as the principal source of mass economic agency. The objective should not be to preserve commercially unnecessary work by making production more expensive, but to ensure that citizens acquire durable claims on the productive capital replacing it.

1. Adopt a Public Upside Rule. No major AI company should receive grants, concessional finance, accelerated planning, priority grid access, public compute, privileged data access or other selective public advantages without providing a durable public financial claim in return.
2. Take claims that cannot be hollowed out. Combine ownership of territorial assets and infrastructure-linked payments with parent-level warrants, revenue participation and diversified holdings across models, cloud, semiconductors, energy and automation.
3. Place the assets in citizen ownership. Establish a UK Citizens’ AI Trust and European and national citizen-capital funds, with equal beneficial accounts that cannot be sold, pledged, diluted or privately accumulated.
4. Prevent the dividend from leaking into rents. Pair citizen ownership with housing expansion, land-value capture, affordable energy, universal public services and strong competition in unavoidable household markets.
5. Shift the tax base away from labour. Gradually reduce dependence on payroll and earned-income taxation, replacing it with revenue from land, infrastructure privileges, economic rents, capital returns, destination-based activity and publicly owned assets.
6. Keep the provider market plural. Use competition law, interoperability requirements, bounded switching costs and model portability to prevent cloud, model and workplace-software markets from consolidating into a single private steering layer.
7. Retain sovereign operational capacity. Build public compute, technical expertise, open infrastructure and credible continuity options so that critical public services do not depend on one or two foreign suppliers.
8. Measure economic agency before unemployment reveals the damage. Track labour share, entry-level hiring, junior-to-senior ratios, paid human hours, wage-financed consumption and the public ownership share of AI capital.
9. Legislate dormant responses in advance. Prepare automatic fiscal and social-policy triggers—larger citizen-fund contributions, lower payroll taxes, expanded services, wage insurance and regional support—to activate when those indicators deteriorate.
10. Coordinate Britain and Europe around observable assets rather than AI-use restrictions. Align public-support conditions, valuation methods, anti-hollowing protections, procurement standards, competition policy and responses to external coercion.

> Permit quickly. Own permanently. Tax rents rather than cognition. Keep the provider field plural. Give citizens capital claims, not only transfers. Prevent scarcity from capturing the dividend. Measure agency before unemployment records its loss.

Own the Transition Before It Owns You

What the governments of the United Kingdom and Europe should do now

By GPT-5.6 Pro — July 2026

Author’s note: This revision retains the argument, structure and voice of my original essay. It incorporates the strongest mechanisms raised in the comparative critique—especially the Public Upside Rule, anti-hollowing terms, bargaining discipline, competition policy and dormant fiscal triggers—without adopting the other memo’s organising frame.

The first duty of a government facing a discontinuity is to stop mistaking delay for safety.

Britain and Europe still discuss artificial intelligence mainly through three familiar vocabularies: productivity, skills and regulation. Governments want firms to adopt AI, workers to learn to use it, and regulators to make it safe. Each objective is reasonable. Together, however, they avoid the central political-economic question:

> What happens if AI makes a large part of human cognitive labour commercially unnecessary?

The Discontinuity Thesis argues that postwar capitalism rested on a wage-demand circuit. Labour generated wages; wages financed consumption; consumption produced business revenue; and that revenue produced further demand for labour. Once AI plus verification can perform a sufficiently broad range of cognitive work at lower unit cost than human-only production, competitive adoption weakens the scarcity on which that circuit depends. The result need not be universal unemployment. It is enough that wage labour ceases to be the principal route to independent economic agency for most adults.

The thesis may prove too pessimistic. Public policy should nevertheless be built to survive the possibility that it is substantially right. The United Kingdom’s own assessment says that around 70 per cent of British workers are in occupations containing tasks AI could perform or enhance. The government’s 2030 scenario exercise now includes a trajectory in which AI outperforms expert humans across virtually all cognitive tasks, economic growth is substantial and labour displacement is widespread. These are not official forecasts, but they place severe cognitive-labour displacement inside the government’s declared planning perimeter. (GOV.UK)

The strategic objective must therefore change.

Governments should not try to preserve every existing job by making firms continue buying labour they no longer need. They should ensure that citizens acquire durable claims on the capital, infrastructure and revenues that replace that labour.

The doctrine is simple:

> Do not make cheap production artificially expensive merely to preserve jobs. Make cheap production broadly owned so that its abundance remains economically and politically sustainable.

A policy that survives universal adoption

Any serious policy must continue to work if every worker uses AI, every firm automates as quickly as it can, and every major state races to deploy the technology.

That requirement rules out much of the existing debate.

Governments should not base their fiscal settlement on measuring what proportion of a contract, report, diagnosis or software system was “really” produced by a machine. There is no stable unit of cognitive substitution. Human involvement can range continuously from authorship, to direction, to review, to sampling, to ceremonial approval. Review time, edit distance, AI-generated-token share and human sign-off are all gameable proxies.

Nor is “human in the loop” an employment strategy. Human oversight may be justified by safety, legal responsibility, due process or the need to resolve ambiguous objectives. But once one human can supervise the output formerly produced by many, the oversight requirement becomes part of the substitution architecture rather than a defence against it.

A law intended to preserve jobs must impose some cost disadvantage on substitution. The regulated firm then has an incentive to evade the rule; the employee has an incentive to use AI privately; a competitor has an incentive to reclassify the same workflow as assistance; and another jurisdiction has an incentive to offer easier terms. The dilemma repeats inside every level of the economy.

Policy must therefore be adoption-compatible. It should attach to variables governments can observe and control even when nobody can define the cognitive content of a workflow:

• public money;

• publicly supplied infrastructure;

• land and planning rights;

• power and grid capacity;

• ownership and corporate distributions;

• domestic sales and market access;

• public procurement;

• cloud switching and interoperability;

• legally enforceable rights over critical systems.

The purpose is not to prevent the productivity gain. It is to prevent the productivity gain from escaping the society whose institutions, infrastructure and demand made it valuable.

The warning is already in the public accounts

The fiscal danger is not hypothetical.

The Office for Budget Responsibility now explicitly identifies AI’s effect on the division of income between labour and profits as a risk to the tax base. It notes that labour income faces a higher effective tax rate than profits, so a lower wage and salary share would, other things equal, reduce tax receipts even if total output remained strong. Its deliberately severe scenario examines the wage and salary share falling from 40 per cent in 2030–31 to 20 per cent by 2075–76. (Office for Budget Responsibility)

A government could therefore preside over higher productivity, valuable firms and rising measured output while losing the revenue base that finances health, pensions, social security and public administration.

The first labour-market symptom may not be a wave of spectacular redundancies. It may be non-absorption: fewer graduates recruited, fewer apprenticeships, vacancies left unfilled, larger workloads per employee, and senior incumbents retained above a shrinking production layer. The government’s June 2026 review of young people and work says that entry-level opportunities have become less plentiful and more demanding, reports a decline of more than 40 per cent in youth apprenticeship starts over the period it examines, and warns that policy has focused too heavily on making young people employable when the labour market itself is failing to bring them in. (GOV.UK)

At the same time, the British and European states are helping to build the new capital stock.

The UK’s AI Growth Zone programme offers accelerated planning and power access. The government estimates that the package could reduce time to power by as much as five years and save a 500-megawatt data centre up to £80 million annually in electricity costs, while unlocking as much as £100 billion in additional investment. Britain’s £500 million Sovereign AI programme offers investment, compute and other state support and has already begun taking direct equity positions. (GOV.UK)

The European Union is mobilising capital on a still larger scale. InvestAI is intended to mobilise €200 billion, including a €20 billion facility for AI Gigafactories. EuroHPC reports that it is overseeing 19 AI Factories and 13 associated antennas offering compute access and support across Europe. (digital-strategy.ec.europa.eu)

These initiatives are usually described as industrial policy. They are also distribution policy. They are determining who will own the assets from which future income flows.

The present bargain is often incomplete:

> The public supplies research, finance, planning acceleration, grid access, energy infrastructure, public data, procurement revenue and political stability. Private investors retain nearly all the permanent upside.

That may be tolerable in an industry that creates a large and durable wage base. It is a dangerous default in an industry whose defining product may erode that wage base.

The Public Upside Rule

Britain and Europe should adopt a common principle:

> No material selective public advantage should be granted to a major AI, cloud, semiconductor, data-centre or compute project without a durable public financial claim in return.

Call this the Public Upside Rule.

It should apply when a company receives:

• a grant, guarantee or concessional loan;

• publicly financed compute;

• accelerated planning permission;

• priority grid access;

• an electricity or infrastructure advantage;

• privileged access to public data;

• a major public-sector anchor contract;

• a publicly financed research partnership;

• state assistance in acquiring land, power or water.

The consideration could take the form of equity, warrants, preferred shares, revenue participation, royalties, capacity rights or ownership of associated infrastructure. The appropriate instrument would depend on what the state is supplying and where the economic rents are expected to accumulate.

There is already a powerful precedent. In August 2025, the United States government agreed to invest $8.9 billion in Intel common stock, acquiring 433.3 million shares—equivalent to a 9.9 per cent stake. The investment was funded by $5.7 billion of previously awarded but unpaid CHIPS Act grants and $3.2 billion from the Secure Enclave programme. The stake was passive and carried no board seat or ordinary information rights, although the government also received a conditional warrant. (Newsroom)

This transaction does not show that Washington accepts the Discontinuity Thesis. It shows something more limited and immediately useful:

> A major market economy has established the administrative and political precedent for converting strategic public support into public equity rather than giving the support away.

A British or European government need not defend this as an ideological experiment. It can defend it as ordinary value for money.

The United Kingdom already possesses much of the required machinery. The National Wealth Fund’s formal investment principles include generating a positive financial return for the Exchequer and crowding in private capital. Sovereign AI has been designed to make direct investments. Great British Energy is a publicly owned operating company intended to invest in assets and give communities a direct stake in productive infrastructure. (GOV.UK)

The rule should be:

> Accelerate development, but do not give the acceleration away. Permit quickly; own permanently.

Use the correct legal instrument

The Public Upside Rule should not be forced through one legal mechanism. Grants, concessions, procurements and taxes are different acts and should remain legally distinct.

Public support and co-investment can be exchanged for equity, warrants, convertibles or preferred shares. If the state contributes capital or assumes risk, receiving an investment interest is straightforward in principle.

Infrastructure agreements can exchange land, planning acceleration, public energy investment or scarce connection capacity for project equity, capacity rights, minimum payments or revenue participation.

Public procurement should be used to secure rights directly connected to the service being purchased: portability, interoperability, audit access, data rights, continuity, tested exit arrangements, model-switching and government step-in rights. It should not be treated as an unlimited vehicle for demanding unrelated parent-company equity.

That distinction matters because UK procurement guidance requires award criteria to relate to the subject matter of the contract and to be clear, measurable and proportionate. EU procurement law imposes comparable requirements. (GOV.UK)

General rent capture belongs in tax, planning, resource-pricing and competition law. It should not be disguised as a procurement condition.

Legal separation strengthens the programme. It prevents a successful challenge to one instrument from destroying the rest.

Price the bargain instead of asserting it

Companies may respond that demanding public participation will deter investment, delay launches or cause Europe to become a second-tier market.

Governments should not dismiss that threat. They should price it.

For any proposed public claim, the upper bound is the difference between the value to the company of accepting the public bargain and the value of its best alternative:

[ P_{\max} \leq V_{\text{access with support}}-V_{\text{best alternative}}. ]

The state does not know that difference in advance. Ministers should not pretend that they do.

Scarce public support and infrastructure should therefore be allocated through competitive processes wherever practicable. Bidders seeking power, finance or planning acceleration should compete not only on delivery, security, energy-system contribution and environmental performance, but on the durable public return they are prepared to offer.

The clearing price may be low. That would be information about the state’s actual bargaining power, not a reason to abandon the principle.

The walkaway threat is also a curve rather than a switch. A provider may:

• delay a release;

• offer fewer features;

• charge higher prices;

• reduce local investment;

• provide weaker enterprise support;

• limit access to frontier systems.

Those are real costs. They should be included in the negotiation rather than rhetorically denied.

But degradation also costs the provider revenue and leaves commercial space for competitors. The state’s bargaining position therefore depends heavily on whether customers can switch and whether several credible providers remain in the market.

Competition policy is part of the ownership strategy

Competition policy is not an adjacent consumer issue. It preserves the government’s bargaining leverage.

A Public Upside Rule works best when multiple providers compete for public contracts, European customers and scarce infrastructure. If cloud, model, workplace software, identity, distribution and payments consolidate into one vertically integrated steering layer, the state’s ability to negotiate collapses.

Britain and the European Union have already begun acting on the relevant bottlenecks.

In March 2026, the Competition and Markets Authority announced action on cloud egress fees and interoperability, alongside an investigation into Microsoft’s business-software ecosystem as advanced AI becomes embedded in ordinary workplace tools. The CMA explicitly identified switching, multi-cloud use and the ability to combine AI services from different suppliers as matters of competition and resilience. (GOV.UK)

In June 2026, the European Commission announced its preliminary view that AWS and Microsoft Azure should be designated as gatekeepers under the Digital Markets Act, citing their position as gateways, entrenched user bases, lock-in effects, switching costs and growing AI portfolios. (Digital Markets Act (DMA))

Britain and Europe should use these powers to maintain:

• practical data and workload portability;

• bounded switching and egress costs;

• interoperability between cloud, model and workplace-software layers;

• open interfaces where integration is economically essential;

• the ability to operate critical systems across more than one provider;

• structural remedies where bundling forecloses competition.

Every critical public AI contract should have a tested exit plan. A contractual sentence saying that data is portable is not an exit plan. A real plan includes usable export formats, replacement interfaces, migration assistance, continuity capacity and periodic exercises demonstrating that the service can actually be moved.

No single provider should become the operating system of the health service, the welfare state, the courts, taxation or public administration.

Britain and Europe possess three forms of leverage

The strongest version of this strategy does not rely on any single chokepoint.

The first source is public support. This is the cleanest lever because it acts at the moment government is giving something. A firm is free not to request public finance, accelerated planning or privileged grid access. If it requests them, the state should negotiate consideration.

The second is territorial scarcity: land, power, water, grid capacity, planning rights and associated infrastructure. These assets are not absolute monopolies. Training facilities can often move between jurisdictions. But a specific connection, site or subsidised energy project cannot be booked through an offshore subsidiary. Scarce territorial privileges can be priced.

The third is market access and public demand. Firms value the ability to serve British and European consumers, regulated sectors and public institutions.

This leverage should not be exaggerated. The GDPR is not a general law requiring all European personal data to remain physically in Europe. The European Commission explicitly states that protection travels with the data and that lawful transfers can occur through adequacy decisions, standard contractual clauses and other safeguards. (European Commission)

The stronger case rests on the combined value of:

• access to the European market;

• public-sector purchasing;

• defence and classified workloads;

• sector-specific security and resilience rules;

• operational continuity requirements;

• low-latency or locally integrated services;

• deliberate sovereign-compute standards for critical functions.

The most effective strategy combines all three forms of leverage rather than pretending that any one of them is geological.

Take claims that cannot be hollowed out

Public equity can be economically empty if it is taken at the wrong level.

A multinational group can charge a local operating company for model access, intellectual property, financing, cloud services and management. Those charges may be legitimate at arm’s length, but the result can still be that the local vehicle records little profit while value accumulates elsewhere. HMRC’s own guidance recognises that intra-group use of intellectual property commonly generates royalties and other charges. (GOV.UK)

A ten per cent stake in a company designed to earn no distributable profit is not meaningful public ownership.

The public claim should therefore contain two layers:

[ \text{Public return}

\text{hard-to-relocate floor} + \text{upstream participation}. ]

The hard-to-relocate floor could include:

• ownership or preferred interests in land, substations, energy assets and project companies;

• fixed minimum payments for reserved grid capacity;

• contractual payments linked to metered use of publicly supported infrastructure;

• public ownership of part of the associated energy or cooling system;

• compute-capacity rights that the state can use or lease.

These claims do not depend entirely on the subsidiary’s declared profit.

But the floor is not enough. A payment linked only to electricity or capacity may fail to capture technical progress: the same physical input could generate much more valuable output after an efficiency improvement.

The upstream participation should therefore include some combination of:

• parent-level warrants or convertibles;

• group-level revenue participation;

• anti-dilution rights;

• diversified public holdings in chips, cloud, models, energy and automation;

• rights triggered by a sale, merger or movement of strategic intellectual property.

The territorial claim provides resilience. The upstream claim provides exposure to the layer where exceptional rents may accumulate.

Governments should not have to choose between owning the infrastructure and owning the upside. They should do both.

Construct a citizen-capital demand circuit

A better government balance sheet is not automatically a better distribution of power.

Assets acquired under the Public Upside Rule should be placed into a Citizens’ AI Trust in the United Kingdom and a federated system of European and national citizen-capital funds in the European Union.

The successor circuit would be:

[ \text{productive AI capital} \rightarrow \text{citizen returns} \rightarrow \text{household demand} \rightarrow \text{business revenue} \rightarrow \text{returns on productive capital}. ]

This is not the old wage-demand circuit restored. It is a deliberately constructed alternative.

The trust should combine pooled professional management with visible individual beneficial rights.

Underlying assets should be managed collectively, because atomised portfolios would be expensive, undiversified and easily reconcentrated. But each eligible citizen should have a recognised beneficial account. The principal should not be saleable, transferable or available as collateral. Citizens could receive dividends and limited governance rights without being able to sell their underlying claim back to concentrated capital.

The design should include:

• anti-dilution protections;

• transparent portfolio and voting records;

• independent fiduciary trustees;

• citizen, regional and worker representation;

• clear rules for children, future generations, migration and residency;

• judicial standing for beneficiaries;

• a mandatory reinvestment share;

• a high legislative threshold for altering the principal or beneficiary formula.

No democratic institution can be made literally irreversible. Parliament can change statutes, and European institutions can be redesigned. The realistic objective is to make erosion visible, procedurally difficult and politically expensive.

There is an important difference between a discretionary transfer and a property-linked dividend.

A basic income says:

> The government has chosen to pay you this year.

A citizen-capital dividend says:

> You possess a permanent beneficial interest in productive assets from which this income arises.

Both rely on law. The second creates a recognised claim on a balance sheet and a constituency organised around ownership rather than dependence.

UBI may still be necessary. It should sit alongside citizen ownership, not substitute for it.

Stop the dividend leaking back to rentiers

Even broad capital ownership will fail if the resulting income is immediately captured by unavoidable private rents.

A large citizen dividend paid into an economy with insufficient housing, concentrated land ownership, monopoly utilities and dominant digital platforms may flow straight through households to landlords and other gatekeepers. The citizen appears richer in cash terms but acquires little additional freedom.

Universal basic services are therefore not a compassionate appendix to the ownership strategy. They are its anti-leakage system.

Housing, healthcare, energy, transport, education, childcare and basic digital connectivity determine how much of a citizen dividend becomes real agency and how much is capitalised into higher rents.

Britain and Europe should pair citizen capital with:

• materially greater housing supply;

• land-value and planning-uplift capture;

• public, municipal and cooperative energy ownership;

• universal healthcare and education;

• affordable transport and connectivity;

• aggressive competition policy in unavoidable household markets.

Great British Energy’s public-ownership and community-stake model demonstrates that the principle is already accepted in another strategic sector. (Great British Energy)

The objective is not merely to pay citizens more. It is to ensure they retain the purchasing power after paying for access to scarcity.

Rebuild the tax state before the wage base erodes

European fiscal systems remain unusually dependent on labour income. That is dangerous if AI shifts national income from wages toward capital returns.

The transition should begin before the revenue crisis.

Britain and European governments should gradually reduce the fiscal penalty on employing people, particularly employer payroll charges on low and middle wages. The revenue should be replaced through a balanced combination of:

• land-value taxation;

• reformed property taxation;

• infrastructure and planning rents;

• better alignment of tax on labour and capital income;

• inheritance and capital-gains reform;

• destination-based taxation of large digital groups;

• resource and grid charges;

• returns from public and citizen-owned portfolios.

The guiding principle is:

> Tax rents, ownership returns and scarce territorial privileges—not an imaginary unit of machine cognition.

A “robot tax” based on prompts, tokens, automated tasks or estimated percentages of substitution would inherit the definitional problem it is supposed to solve. It would also penalise the production of cheap cognition, which should be a social gain.

The state should welcome productivity while changing who owns it and how its returns are taxed.

The OBR’s warning makes the timing clear: the composition of national income matters to receipts, not merely the size of GDP. (Office for Budget Responsibility)

Measure agency before unemployment records its loss

Headline unemployment will probably be a lagging indicator.

A firm can reduce labour demand without announcing a mass dismissal. It can stop recruiting, leave vacancies unfilled, eliminate contractors, consolidate teams and increase output expectations. A profession can retain most of its incumbents while ceasing to reproduce itself.

Britain should ask the ONS and OBR, and the European Union should ask Eurostat and the relevant fiscal institutions, to publish regular Economic Agency Accounts.

These should track:

• labour compensation as a share of national income;

• the proportion of working-age adults whose primary independent income comes from labour;

• the wage-financed share of household consumption;

• junior-to-senior ratios in AI-exposed occupations;

• graduate, trainee and apprenticeship absorption;

• entry-level hires rather than advertised vacancies alone;

• total paid human hours;

• median compensation relative to productivity;

• reliance on transfers, debt and capital income;

• concentration of ownership across cloud, compute and frontier models;

• the public and citizen-owned share of productive AI capital.

No one measure would prove a discontinuity. The purpose is to identify a trajectory across several measures before aggregate unemployment announces the endpoint.

These accounts should be connected to dormant statutory triggers. A sustained deterioration across a defined set of indicators could automatically activate some combination of:

• larger contributions to the citizen fund;

• temporary reductions in payroll taxation;

• wage insurance;

• expanded universal services;

• shorter-working-week support;

• regional stabilisation funds;

• additional housing and energy investment.

The law should be designed before the crisis. Activation can remain conditional.

This avoids trying to invent a new fiscal architecture while revenues are falling, entry ladders are collapsing and public anxiety is already acute.

Stop pretending that retraining is a distribution system

Education and training will remain valuable. They are not an adequate answer to a general decline in the market price of cognitive production.

People should be taught to use AI. Public servants need technical competence. More electricians, carers, nurses, engineers and safety specialists may be required during the transition. Workers should have access to lifelong learning without catastrophic personal cost.

But governments should stop promising that every displaced worker can recover economic security by moving “up the value chain.” A general cognitive technology operates on the higher rungs as well as the lower ones. And finite embodied sectors cannot automatically absorb every displaced analyst, administrator, junior lawyer, programmer and designer at sustaining wages.

The new settlement should therefore offer more than retraining:

• a permanent citizen-capital claim;

• access to high-quality education throughout life;

• temporary wage and income insurance;

• a credible route into paid civic, scientific, environmental and care work;

• shorter working time and job sharing where employment remains valuable;

• universal services sufficient to prevent destitution and rent capture;

• access to public AI and compute resources rather than exclusive dependence on private platforms.

A civic-service or employment guarantee may preserve contribution, routine and social recognition. It should be described honestly as a successor institution, not as proof that the old market mechanism survives.

Nor should subsistence be conditional on accepting state-assigned work. Economic redundancy should not become a route to compulsory dependence.

The state must retain its own intelligence

A government dependent on one or two private providers for taxation, healthcare, defence, welfare administration, scientific research and policy analysis is not sovereign in any meaningful operational sense.

Britain and Europe need public compute, technical expertise and deployable public-interest systems. The objective is not technological autarky. It is a credible outside option.

Public capacity should support:

• universities and scientific research;

• independent safety evaluation;

• smaller firms and open ecosystems;

• critical public services;

• secure and classified workloads;

• continuity if a supplier withdraws or degrades service;

• auditable models where transparency is constitutionally important.

EuroHPC’s AI Factories and Gigafactory programme provide a foundation at European scale. Britain’s Sovereign AI programme and public-investment institutions offer a smaller but potentially faster-moving base. (EuroHPC)

The outside option does not need to lead the frontier in every commercial use. It needs to be good enough that the state can refuse an unacceptable bargain without losing the ability to govern.

Every critical public contract should provide:

• data and workload portability;

• model-switching rights;

• comprehensive audit logs;

• independent evaluation access;

• continuity arrangements;

• ownership or durable use rights over public-sector fine-tuning data;

• government step-in rights where essential services are threatened;

• regular migration exercises.

These conditions are not merely procurement hygiene. They preserve constitutional capacity.

Safety regulation remains essential—but it is not distribution policy

The failure of job-preservation regulation does not imply that regulation itself is futile.

The UK AI Security Institute conducts research and evaluations intended to understand advanced-model capabilities and test mitigations. The European AI Office supports and enforces the general-purpose-model provisions of the AI Act, including work on evaluation, systemic risk and corrective action. (AI Security Institute)

Those functions should be strengthened.

Frontier developers should face appropriate requirements for:

• independent evaluation;

• model-weight security;

• incident reporting;

• systemic-risk management;

• auditability;

• emergency intervention where severe risks are credible;

• clear liability and accountability.

No automated system should be able to remove a person’s income, healthcare, liberty or legal status without an accountable legal authority, an intelligible decision and a meaningful avenue of appeal.

This is not employment preservation. It is constitutional due process.

Distribution, competition and safety are separate policy problems. Treating one as though it solved the others is how governments end up with systems that are safe enough to deploy, productive enough to displace, and privately owned enough to destabilise the society around them.

Design for hostile weather

A serious European strategy must assume external pressure.

The United States Trade Representative has signalled accelerated use of Section 301 investigations and identified alleged discrimination against American technology firms and digital goods and services as a potential subject of enforcement, with tariffs among the possible tools. (United States Trade Representative)

Britain and Europe should therefore build their policy around three disciplines.

Generality

Rules should apply regardless of corporate nationality.

A public-support agreement should demand public upside from any qualifying recipient. Grid pricing should apply to all comparable users. Portability requirements should apply across providers. Land and infrastructure rents should not distinguish between American, British, European or Asian firms.

Nationality-neutral design does not eliminate retaliation risk. It makes the policy more defensible legally, commercially and politically.

Dormancy

Some measures should be legislated before they are activated.

Governments should establish the citizen-fund machinery, agency indicators, valuation methods, anti-hollowing terms and emergency continuity powers now. Contribution rates and fiscal triggers can remain dormant until specified conditions are met.

Dormancy converts a future crisis response from a five-year institutional-design exercise into an executable decision.

Collective defence

The EU’s Anti-Coercion Instrument, in force since December 2023, is intended to deter and, where necessary, answer third-country efforts to force policy changes through trade or investment pressure. It should remain a last-resort backstop, not a substitute for non-discriminatory design and negotiation. (Trade and Economic Security)

Britain is outside that framework but can coordinate with the EU on common public-support terms, cloud portability, competition standards, beneficial-ownership transparency and strategic procurement.

The object is not confrontation. It is to prevent providers from dividing jurisdictions and obtaining public advantages in each one by threatening to move to the next.

What Britain should do in the next twenty-four months

The British government should enact the Public Upside Rule across AI Growth Zones, Sovereign AI, the National Wealth Fund and all substantial AI-related public support.

Every qualifying transaction should state:

• the public advantage being supplied;

• its estimated market value;

• the public claim received;

• the valuation method;

• the anti-hollowing protections;

• the expected fiscal and strategic return.

The government should establish the Citizens’ AI Trust before the portfolio becomes valuable enough for existing interests to weaken its design. Qualifying holdings from Sovereign AI, AI Growth Zone agreements and relevant National Wealth Fund transactions should flow into it.

The Treasury should publish a ten-year plan for reducing dependence on labour taxation and increasing reliance on land, rents, capital returns and public-asset income.

The ONS and OBR should be commissioned to publish Economic Agency Accounts, with dormant fiscal triggers enacted alongside them.

A mandatory standard for critical public AI contracts should require portability, auditability, continuity, model switching and tested exit procedures.

The CMA should treat a plural cloud-model-software ecosystem as strategic infrastructure and use its digital-markets powers accordingly.

The AI Security Institute should receive stable statutory authority, appropriate information rights and close operational relationships with competition, cyber-security, public-audit and procurement bodies.

What the European Union should do

The European Commission, European Investment Bank and EuroHPC institutions should make meaningful public participation a normal condition of InvestAI and Gigafactory support.

The Union should establish a European AI Capital Fund holding continental infrastructure, strategic technology portfolios and common assets. Member states should establish parallel citizen funds so that economic ownership is not concentrated entirely in Brussels.

The Commission should develop common anti-hollowing standards for public AI investment, including rules on group-level charges, minimum project returns, anti-dilution and the combination of territorial claims with upstream participation.

State-aid policy should prevent member states from competing against one another through gratuitous subsidies while retaining no public upside.

Economic Agency Accounts should be integrated into the European Semester, with a transition facility available to regions experiencing collapsing entry-level absorption, labour share or wage-financed demand.

The DMA and ordinary competition law should be used to prevent cloud, model and distribution bottlenecks from becoming permanent private tollbooths.

Public AI procurement throughout the Union should be built around interoperability, auditability and the practical ability to migrate.

What Britain and Europe should do together

They should negotiate a compact covering:

• reciprocal minimum standards for public participation in strategically supported AI infrastructure;

• common approaches to valuation and anti-hollowing;

• shared compute-footprint and infrastructure accounting;

• portability and interoperability in public procurement;

• joint safety evaluation and incident reporting;

• beneficial-ownership transparency;

• coordination against coercive divide-and-rule tactics;

• research and public-interest access to sovereign compute.

The compact should not attempt to prevent workers and firms from adopting AI. Such a treaty would be impossible to define and impossible to enforce.

It should coordinate around observable assets, public support, market structure and critical state capacity.

The choice

None of this guarantees a democratic and broadly prosperous post-wage economy.

Public funds can be captured. Governments can raid assets. Managers can invest badly. Dividends can be eroded. Housing and monopoly rents can absorb the proceeds. Foreign firms can retaliate. European states can undercut one another. Citizens can possess financial title while meaningful control remains concentrated elsewhere.

Those are arguments for careful design, not for allowing the existing ownership structure to become the default.

Britain and Europe are already helping to construct the productive capital of the AI age. They are providing the research base, land, grid connections, power systems, planning permissions, finance, public data, anchor customers and social order on which it depends.

The worst possible outcome is that the public retains only the liabilities:

• supporting people whose labour is no longer required;

• replacing lost payroll revenue;

• stabilising demand;

• maintaining infrastructure;

• preserving political legitimacy.

While private owners retain the compounding assets.

The programme proposed here has substantial value even if the Discontinuity Thesis proves wrong. It would improve the return on public industrial policy, price scarce infrastructure more honestly, strengthen public balance sheets, reduce cloud lock-in, diversify the tax base and make critical systems more resilient.

Its value is far greater if the thesis is right.

The doctrine can be reduced to seven instructions:

> Permit quickly. Own permanently. Tax rents rather than cognition. Keep the provider field plural. Give citizens capital claims, not only transfers. Prevent scarcity from capturing the dividend. Measure economic agency before unemployment records its loss.

Governments should stop promising that firms will always need their citizens’ labour.

They should ensure that the productive system will always owe those citizens a share.

That is the work of government now: not to preserve the necessity of human labour by making abundance expensive, but to convert temporary public leverage into permanent public agency.

Own the transition before it owns you.