Regulating Robots by Owning Them

Two problems are approaching, they are usually treated as unrelated, and the instruments proposed for each are incapable of solving the other. The first is displacement. Robotic systems capable of substituting for human labour across a widening range of tasks are being produced at falling cost. The policy debate about what to do has been almost entirely fiscal: tax the substitution, subsidise the retraining, redistribute the gain. The canonical proposal asks whether robots should pay something equivalent to payroll tax, and the literature that has grown around it argues about rates, bases and incidence. The second is modification. A robotic system in private hands can be altered—in its hardware, in the software it runs, or in the constraints imposed on either—and a system altered outside its design envelope is dangerous in ways its manufacturer did not contemplate and its owner may not understand. The instrument approaching here is the European Union's Machinery Regulation, which applies from 20 January 2027 and, for the first time, addresses tampering directly: machinery must be designed so that connecting another device does not create a hazard, and control systems and software must be secured against unauthorised modification. Neither instrument can solve the other's problem, and the reasons are structural rather than remediable. A tax is a price instrument. It changes what substitution costs and does not control how much of it occurs, so a firm that can absorb the charge proceeds unaffected. It also confers no right of access: taxing a thing gives the state no entitlement to look inside it, which is why no fiscal proposal has ever been offered as an answer to the modification problem. The Machinery Regulation is a market-entry instrument. Its obligations attach to the manufacturer at the moment of placing the product on the market, and they are discharged once. They say nothing about the unit in a citizen's possession three years later running modified firmware. Layered with the Artificial Intelligence Act and the revised Product Liability Directive, the compliance burden on manufacturers is now substantial and it is all front-loaded. Every instrument in the field attaches at the point of sale, and none survives into the operating life. This paper proposes a single structural change that addresses both. Citizens may possess robotic systems; the state retains the property. Possession passes under a lease; title does not. A third capability follows from the same structure, is more consequential than either, and is the reason the paper has to deal with targeting law (Part V §20D). Because title never passes, a state facing armed attack, disaster or epidemic may recall units as the exercise of a reserved right in its own chattel rather than as an expropriation of anyone's property. Expropriation asks whether the state may take what is yours; recall asks whether it may have back what is its own, and the second is a question about the construction of an agreement rather than a hard question answered slowly. The effect is a latent reserve funded by private consumption: a state that subsidises the machines, buys them as a monopsony, and retains their residual value obtains at a fraction of procurement cost a capability its possessors purchase and house. The precedents are old and unexotic—the Civil Reserve Air Fleet has committed privately owned airliners to military airlift since 1951, and forty-five merchant ships were taken up from trade for the Falklands in 1982. That is a defence argument and this paper makes it in terms rather than leaving it to be inferred, because stating it is what exposes the objection that nearly defeats it. Military objectives include objects which by their purpose—meaning intended future use—make an effective contribution to military action. A statute reserving every unit of a designated class for defence tasks in an emergency is a standing statement of intended future use over every unit in the class. On that reading the scheme would confer a military purpose on several million domestic machines, and on the dwellings containing them, before any emergency and without anyone doing anything. The paper's answer is a drafting discipline rather than a denial: the permitted tasks are listed exhaustively and confined to civil-protection functions; munitions carriage, military engineering and any use of a weapon are excluded expressly; recalled units must be delivered away from dwellings before use; and a state wanting the excluded capability must obtain it through a separate designated class that may not be kept at a dwelling. What cannot be drafted away is that a belligerent will still make the argument, and the honest conclusion is that a state unwilling to accept that should adopt the scheme without the recall limb, keeping the quantity and inspection capabilities and forgoing the third. Care and assistance units are absolutely exempt throughout, since their removal would create the emergency the recall claimed to answer. That does two things no other instrument does together. It makes the number of units in circulation a regulated quantity, because a lease may be granted or refused—a labour-market instrument operating on volume rather than on price, and therefore capable of what a tax structurally cannot do. And it preserves a continuing right of access, because inspection is an ordinary incident of a lessor's interest and requires no new regulatory theory to justify. That is what makes any rule against dangerous modification enforceable rather than aspirational. The quantity is graduated, and the graduation is the design. Below a household entitlement, possession is as of right and nothing is rationed. Above it and below an industrial threshold, possession follows an application which states use, place and number and which is designed to be granted rather than refused—so that the tier restricts almost nothing and discloses a great deal. Only at industrial scale does the number bind, and there it is set by reference to identified occupational classes and may be differentiated by sector, region and applicant. Three things follow. The restriction falls where the gain from substitution is captured, so ordinary households are not rationed at all—which a robot tax, falling on the retired person's single unit and the logistics operator's six hundredth alike, cannot manage. The middle tier generates the deployment data from which the binding tier is calibrated, which no fiscal instrument can do, because a tax return records a completed transaction in money while an application states an intention in units, uses and places. And the differentiation is enforceable, because a licence conditioned on use and place is verifiable on the same inspection the modification problem already requires, where a differentiated tax must take the taxpayer's word for the category claimed. The structure is not invented: British goods-vehicle licensing exempts private use, caps the number of vehicles and ties them to a stated operating centre, under a regime that originated in the Transport Act 1968 and has run continuously since 1970. Three consequences for the population are developed at Part V §20B and they are the reason the instrument is politically available at all. The regulated good becomes cheaper, because a lessor retaining the reversion can price below outright sale without loss and a single national channel buys as a monopsony — an unusual combination, since technology regulation ordinarily raises prices and redistribution ordinarily runs through the fiscal system at one remove from the thing distributed. Access can be widened deliberately: the lease term can be lengthened for those who cannot meet a capital outlay, the outright price discounted by identified category, and the free servicing obligation is flat in cash terms and steeply progressive in practice. No fiscal instrument has any of those levers, because a charge reaches the machine and not the person. And timing decides whether the arrangement is a norm or an imposition: legislated before an installed base exists it is a condition of acquisition, legislated afterwards it is the confiscation of an expectation — and coverage, on which all three capabilities depend, is a function of how early the rule attaches and of nothing else. On what a reserve of numbers actually buys, the paper argues from the record rather than from imagination. Rescue robots were deployed at the World Trade Center site within six hours, in the first known use of robots for urban search and rescue; four teams took part. Forty thousand people worked that site, the recovery ran eight and a half months, and more than ninety thousand were exposed to its hazards. The constraint in 2001 was never the concept but the number of machines four teams could carry to Manhattan, because no mechanism existed by which a state could call on machines it did not own. The claim is disciplined in both directions: a domestic machine is not a void-penetration robot and no quantity substitutes for one — but the void search is not where eight and a half months went, and every hour a machine spends on such a site is an hour a person does not. The precedent is not novel and that is the argument's strength. Aviation airworthiness already does exactly this: private ownership, a continuing certificate, mandatory periodic inspection, and compulsory compliance with directives issued long after sale. The paper works that analogy in detail, because it is the one existing system that combines private possession with continuing state access and a functioning inspectorate. The objections are taken at full strength, and the hardest is not the economic one. A right of access sufficient to inspect is a right of access sufficient to surveil, so the instrument that solves the modification problem creates a monitoring infrastructure reaching into private homes. The paper answers structurally rather tha