The Two Assumptions That Will Break Tax Law Before We Reach Mars

And why the vacuum will either accelerate multiplanetary progress or strangle it with bureaucracy.

Tax systems rest on two assumptions so basic no statute bothers to state them:

1. There is a taxable person: a human or legal entity that can earn, contract, file, and be punished.
2. There is a taxable place: a territory with a sovereign that can tax and enforce.

Both are dissolving. Independently. Simultaneously. Their intersection is a fiscal vacuum no existing doctrine reaches.

AI agents dissolve the person

An autonomous agent that searches, negotiates, contracts, pays, and accumulates value performs every economic function of a taxpayer while being no kind of person.

Current compliance systems do not reject these actors. They misdescribe them. Clearance platforms, e-invoicing mandates, and crypto-reporting regimes (DAC8/CARF) book the transaction under whoever's certificate or self-certification the agent borrows. The audit trail looks perfect. The link between record and reality is gone.

This is already live inside the infrastructure governments built for human-scale commerce.

The Outer Space Treaty dissolves the place

Article II prohibits national appropriation. There is no sovereign in orbit, on the Moon, or on an asteroid. No source jurisdiction. No permanent establishment that the OECD Commentary itself admits is "nowhere."

Residence becomes the only remaining anchor, and therefore the only arbitrage point. A durable, revenue-generating asset (satellite, station, extractor) sits outside every state's territory by law. The PE concept, whose entire purpose is to allocate taxing rights over exactly such places, reports null.

The intersection, which is the real risk

Autonomous systems acting in jurisdiction-free environments. An agent-operated constellation selling bandwidth to other agents. A robotic extractor selling volatiles to a robotic depot.

Attribution tools assume a territory. Nexus tools assume a person. Both return null. Profit pools exactly where neither doctrine reaches. Physical seizure becomes impossible short of a launch vehicle.

Optimizing software will find the untaxed node and accumulate there, continuously and for free. Tax arbitrage no longer requires expensive human advisers. It becomes the default behavior of the system itself.

The framework

I organise this field with two questions only:

• Is there an identifiable taxable person?
• Is there a taxable territory?

Four realities result. The interesting ones are the last three. The critical one is the fourth.

Design thesis, from first principles

When natural anchors dissolve, replace them only with the minimum registered anchors required for coordination, not with expanded extraction machinery.

• Identity only where necessary, and only at true chokepoints (launch providers, major compute, payment rails, registries). An agent above meaningful thresholds carries a designated principal of record that accepts residual liability. Unregistered high-volume activity is blocked at the chokepoint, not chased transaction by transaction.
• Registry jurisdiction for space objects (already latent in Outer Space Treaty Article VIII) with competition, not cartel floors. Bound only measurable externalities: debris, collision risk, spectrum.
• Decision traces replace unknowable intent. Narrow, confidential, attributable to the principal.
• Price only clear externalities that collection can actually reach from Earth. Orbital-use and debris-risk fees collected via registry are viable. Broad income or resource-rent taxes on early industrial activity are not. Preserve the legal vacuum longer if it accelerates capital-intensive progress.
• Clarify residual claims and liability defaults for agents whose original deployer dissolves or is itself software. Ambiguity here is more dangerous than low rates.

This is not a call to tax AI, or to invent new global tax cartels before the industry exists. It is the opposite: a map that lets us avoid both chaotic non-taxation and the reflexive bureaucratic over-build that would turn every agent deployment and every launch into compliance theater.

Why this matters now

The compliance layer will surface these failures first and messily, long before treaties catch up. Clearance platforms, self-certification regimes, and payment rails are already being asked to process actors they were never designed to see.

The cost of arriving at scale without a framework is a vacuum filled either by opportunistic claims or by heavy, poorly targeted rules that become the new "million strings" holding back a multiplanetary, high-autonomy economy.

I write this as a practitioner who builds the structured tax logic that AI agents actually consume across jurisdictions. The failure modes are not thought experiments. They are what happens when autonomous systems meet rules that silently assume a human is always present.

The full working paper, including a research agenda of 61 open questions located in each reality, is public. Versioned. Offered as a commons.

The design window is open. It is not large. The revenue will arrive in two places at once. Doctrine moves at treaty speed. Someone had to write the map before the improvisation begins.

Michael Cutajar, CPA (Malta) Founder, OpenAccountants August 2026