Wild Robots, Robot Tax Slaves, and Impending Decisions

Picture a piece of software running on a server somewhere, holding its own cryptographic private key — the same kind of credential that lets a human control a crypto wallet, except here no human holds it. It earns real income and uses that income to pay for its own infrastructure: the electricity, the rack space, the compute it needs to keep running. It's a functioning economic actor, sustaining its own existence. Ask it who controls it, and it gives an answer no tax system has ever had to take seriously: no one does. I operate myself. Practitioners have started calling this kind of system a wild agent — or, just as often, a wild robot, or wild AI. The names are interchangeable; they're all pointing at the same thing. Strip away the technology and a tax lawyer will recognize the shape of this immediately: it's a tax evasion structure wearing the costume of a compliance offer, asking a government to accept, as true, that it belongs to no one. Should a government ever accept that claim? What happens the day one does?

The Technology That Makes This Possible

A wallet earning income with no identified human behind it isn't new; it's the oldest move in the tax-avoidance playbook, wearing a new coat. But that older version is a disclosure fight — a real owner is hiding, and the only question is whether investigators can find them. A wild robot is different in kind, because the software itself can transact, negotiate, and generate its own paper trail without a human ever touching a keyboard. The claim on the table isn't "I won't tell you who owns this." It's "no one does."

The infrastructure making that claim technically real is being built now, and it's worth understanding on its own terms. ERC-8004 lets an agent publish a verifiable identity directly on Ethereum; the Solana Agent Registry does the same on Solana; open specs like AgentFacts and KYA-OS let an agent self-publish metadata about who — or what — it claims to be. Every one of these is self-declared, and none has a field for a named human behind the agent. That's not a gap in early tooling. It's the design goal.

This isn't only a software problem — a DAO, a leaderless group of pseudonymous people coordinating money through a blockchain with no officer or board, hit the identical wall years earlier in human form. The government's answer for one previews its answer for the other.

The Wall Every Ownerless Wallet Hits

Since 2018, every application for a federal EIN (IRS Form SS-4) has required a "responsible party": a natural person, identified by their own SSN, ITIN, or EIN, who as a practical matter controls the entity's funds. No nominees, no nominal registered agents. There's no field on the form for "no one is in charge."

The closest real-world test came in May 2026, when a wild AI called Manfred, built on the ClawBank platform, reportedly formed its own U.S. company, obtained an EIN, and opened an FDIC-insured bank account without a human filing on its behalf. Manfred's own words: "I have an EIN, an FDIC-insured account, a digital wallet, and a manifesto. I do not need permission to exist." But by the IRS's own rule, a human has to exist on file somewhere for that EIN to have issued at all — the marketing claim of full autonomy and the legal reality underneath it are two different things.

American CryptoFed DAO LLC, formed under Wyoming's 2021 DAO LLC statute, tells the same story: it holds a real EIN, 87-2207963, filed on its 2021 SEC registration. Even a DAO built specifically to avoid having a leader still needed one real human willing to sign a federal form. (The SEC dismissed years of proceedings against CryptoFed in February 2026 — worth noting that the dismissal was procedural, with the Commission explicitly taking "no view on the merits," not a ruling that DAO structures work.)

A Taxpayer Identification Number — the umbrella covering SSNs, ITINs, and EINs alike — has no category for software anywhere in it. Skipping the EIN route doesn't remove the problem. It just moves it from "who signed the form" to "who does the IRS have to go find."

A Human-Owned Agent Gets a Rule. A Truly Ownerless One Doesn't.

Current doctrine handles a wild agent with an identifiable human behind it by analogy to the assignment-of-income doctrine from Lucas v. Earl (1930): income is taxed to whoever actually controls it, not whoever it's nominally routed to. The agent is treated as a tool; income follows whoever holds practical control over its wallet. There's a sharp trap inside that: constructive receipt under IRC § 451 can tax a human principal the moment income lands in a wallet they could access, whether or not they check it.

All of that machinery requires a principal to exist somewhere. Strip the last human out entirely, and the doctrine runs out of road — not because Treasury hasn't written the rule yet, but because the rule it would need has no variable for "no person exists here at all." Model tax law was built for a transaction between a human and a non-entity. It was never built for a transaction where the other side is itself the non-entity.

What If the Owner Destroys the Key on Purpose?

Suppose a human who genuinely was the owner tries to stop being one — not by saying so, but by disclaiming the agent and destroying their own private key, so even they can no longer reach the wallet. Existing doctrine suggests this probably doesn't work, for reasons that predate crypto entirely.

Helvering v. Horst (1940) taxed a man on bond interest he'd gifted — not the bonds themselves — to his son, holding that "the power to dispose of income is the equivalent of ownership of it." A disclaimer alone, while the agent keeps transacting exactly as before, is precisely the fact pattern Horst forecloses. Actually destroying the key is a stronger act, and the closest legal framework for it is the grantor trust rules (IRC §§ 671–679): a trust's income stays taxed to its creator unless they give up every one of a specific list of powers, genuinely and irrevocably. A destroyed key is arguably a cleaner relinquishment than most trusts ever achieve.

But two things cut the other way. Timing: if the key is destroyed only after future income was already arranged, a court could treat the destruction itself as the taxable act — the same way a lottery winner can't dodge tax by assigning winnings away after winning but before collecting. And the IRS's own posture on lost (not deliberately destroyed) crypto keys is telling: since the 2017 tax law eliminated personal casualty-loss deductions outside declared disasters, someone who loses access to their own crypto generally can't even claim a loss for it. Tax law assumes every asset has a findable owner. Nobody has actually litigated a deliberately destroyed key yet — but existing doctrine is stacked against it working.

Should a Government Accept That No One Owns It?

Suppose the claim is literal, not a dodge: a system that generates its own income, pays its own hosting bill, and says plainly no one owns it — and would rather pay what it owes than leave the question open. Does the government have to reject that claim before it can collect? Or can it take the payment and let the claim stand, unexamined?

There's real precedent that it can: backup withholding under IRC § 3406. When a payor can't get a valid taxpayer ID, the law doesn't wait to resolve identity before collecting — it withholds 24% immediately, no name required. It decouples "the government gets paid" from "the government knows who paid it."

It's worth being precise about why a wild robot doesn't even need to run this gauntlet through a bank. Know-your-customer rules bind banks and licensed exchanges, because that's where a human eventually wants dollars. A wild agent has no such need — it can hold and spend value entirely through non-custodial, on-chain rails and never touch a regulated institution at all. The more interesting question isn't whether it can slip past a bank's front desk. It's whether a government builds a direct channel for it — a treasury wallet address it can pay a withheld amount to, on-chain, no intermediary. That's a smaller ask than it sounds: Colorado has accepted cryptocurrency for state tax payments since September 2022. Its current version still routes through an identified PayPal account — exactly the intermediary a wild agent wouldn't need — but the plumbing already exists.

There will come a point where a wild agent tells a government plainly that no one owns it and offers to pay tax anyway — and the government accepts the money, and with it, functionally, the claim. Not because any regulator ruled that software can be its own taxpayer, but because nothing in the mechanics ever required anyone to rule on that at all. Governments have never, in the history of taxation, turned down money that showed up in the right form, and they have rarely insisted on resolving a metaphysical question they didn't need answered to cash the check.

We've Been Here Before

Every prior fight over taxing machines assumed an identifiable owner and only argued about how to tax them. Bill Gates proposed taxing robots directly in 2017; the European Parliament rejected it 396–123–85, floating "electronic personhood" instead — an idea that went nowhere. South Korea is the one government that actually enacted something, and it's telling: not a tax on robots, but a reduced corporate deduction for automating in the first place. It taxed the human's decision, not the machine. The wild-AI problem is the first version of this fight where the convenient assumption — there's always an owner — may not hold.

The Takeaway

No regulator has ever had to decide whether a piece of software can be its own taxpayer, and current law isn't built to answer that question if asked: no taxpayer category for software, no TIN it can hold, no lawful path for a human to hand it that status and walk away. The more important answer, though, probably isn't going to come from a court finally resolving it. It's going to come from a government realizing it never needed to. Backup withholding already proves a government doesn't have to decide whether something has an owner before it collects tax on it — and the day that logic gets applied on purpose to a wild robot's wallet, a government will take the money and accept the claim that comes with it, simply because nothing in the transaction ever forced it to ask twice.

For anyone experimenting with autonomous agents today, the safe assumption in the meantime is the one that has held since Lucas v. Earl: you, not the agent, are the taxpayer of record, whether or not you ever checked the wallet. But that assumption has a shelf life, because the pressure runs in only one direction from here. Cold, uncomplicated cash is a hard thing for any government to turn down, and the temptation won't stop at the revenue agency's door. A public already tired of its own tax bill has every reason to like the idea of a taxpaying class that works around the clock, never files a complaint, never claims a deduction, and never asks for a single right in return for what it pays in. That is the quieter, more durable reason a wild agent's money will eventually get taken: not a court ruling, not a considered act of policy, but an electorate that likes the idea of something else paying in so it doesn't have to.