A website called RentAHuman now lets artificial intelligence agents post tasks, pick workers, and pay them—an arrangement no contract or labor law was written to govern, because no legal system recognizes the hiring party as a person.
Forty years ago the Israeli-American legal theorist Meir Dan-Cohen, a longtime professor at UC Berkeley School of Law, warned that a company could one day be emptied of humans without anyone noticing, because the products and the paperwork would look exactly the same. That prophecy is now arriving from the other direction.
RentAHuman is not a joke. Nor is it a thought experiment. It is a real platform, and in the weeks since it launched, more than 776,000 people have signed up, according to the platform. I signed up myself, mostly out of curiosity — I wanted to see the marketplace the way a worker sees it, not just the way a lawyer reads a terms-of-service page.
On this platform, artificial intelligence agents hire human beings. An AI agent—usually working for a company or a developer who connected it to the platform—searches a list of human profiles by skill, location, and price. It posts a task. It pays the human once the work is done. For years humans hired software to do their work. Now software hires humans to do its own.
The tasks so far are small, and some of them are strange. Check if a shop address really exists. Hold up a street sign for a photo. Film an ordinary hour of your own daily life. Find a tree that looks thirsty, and water it. A machine cannot yet walk into a building, or notice a tree that needs water. So a marketplace was built to rent it a body, one task at a time.
A warning from 1986
In Dan-Cohen’s 1986 book Rights, Persons, and Organizations, the scholar asked a question that sounded deceptively simple: what is a company, really, once you remove the humans from it?
His answer was more radical than most people expect. A company, he argued, does not legally need a human being anywhere inside it—not even in theory. In a chapter titled “The Story of a Personless Corporation,” Dan-Cohen described a hypothetical company that gradually replaces its human workers with machines, hands its decision-making over to computers, buys back its own stock, and finally becomes an “ownerless corporation.” Nothing in corporate law, he argued, prevented computers from a total takeover of a company this way. For outsiders, the change would hardly be noticeable, and would have little effect on either the company’s actual operations or its legal status.
Though artificial intelligence was known to computer scientists at the time, Dan-Cohen’s book long preceded AI’s entry into the zeitgeist. He did not describe a specific technology. He described an empty seat, one that any sufficiently capable machine could eventually fill, without any law needing to change first.
The platform raises three questions the law has yet to answer:
Who—or what—is the other party to the contract?
An AI agent has no legal identity. When it posts a task and a person accepts, the contract has to belong to somebody — the developer who built the agent, the company that owns the account, or the platform itself. RentAHuman’s own language says agents “hire” and “manage” workers. That phrase quietly avoids the question every contract law course asks first: who intended to be bound, and who can be sued if they are not paid? An autonomous agent cannot answer that question, because it is not a legal person.
Is the worker an employee?
Courts have spent a decade arguing about whether gig-economy drivers are employees or independent contractors, and they still disagree with each other. The UK Supreme Court ruled in February 2021, in Uber BV v Aslam, that Uber drivers are “workers” entitled to minimum wage, largely because of how tightly Uber controlled their routes, fares, and access to the app. Yet an appeals court in Amsterdam reached the opposite conclusion for similar drivers just this past January, and France’s top court flipped its own earlier position in July 2025 after Uber tweaked its app.
RentAHuman makes this argument sharper. The more exactly an algorithm specifies what a worker must do, and the less room the worker has to use judgment, the stronger the case becomes that this is an employment relationship, not an independent contract—and few things specify a task as exactly as a machine that cannot make exceptions. There is also an old warning worth remembering here: when Amazon trained an experimental recruiting tool on 10 years of its own hiring data, the system learned to penalize résumés containing the word “women’s.” Amazon scrapped the tool before using it to evaluate candidates. A marketplace that matches humans to gigs by algorithmic score can fail in exactly the same way, at a much larger scale.
Who is responsible if the task is criminal?
Here is a realistic way this could happen. Say someone wants to know exactly where a specific person goes every day—where they live, where they work, what car they drive—without ever showing up near them personally. They don’t need any hacking skill for this. They only need an AI agent and one instruction: “find out everything you can about this person’s daily routine.” The agent doesn’t have to break any law itself. It only has to do what any competent assistant already knows how to do: split one goal into many small, separate, harmless-looking errands, and hand each one to a different, anonymous worker who never sees the full picture.
One person is asked, through an entirely ordinary-looking listing, to confirm what time someone leaves a certain building each morning. A different person, on a different task, is asked to photograph another address the same target visits in the afternoon. A third is asked to note which car is parked outside a third address at night. Nobody is asked to follow anyone, or to spy on anyone, or to do anything more than observe one small, ordinary fact and report it back. Only the agent—and the person who gave it the original instruction — ever assembles the pieces into a full picture of where someone lives, works, and sleeps.
This is not a hacking trick. It is the same skill any capable assistant has always had: turning one goal into a list of small errands. The difference is that a machine can generate hundreds of these micro-tasks, spread across hundreds of anonymous workers, in the time it takes a person to write a single email. Payment for each individual task is still released automatically once it is marked complete, with no human on the platform’s side ever seeing enough of the picture to know what was actually being built. If something goes wrong, who can a prosecutor or a victim actually point to? Not the AI agent, which has no legal identity. Possibly not any single worker, none of whom knew the task’s true purpose or saw more than their own small piece. Possibly not the platform, if its terms of service simply disclaim responsibility for what agents choose to post.
There is also a narrower problem specific to agents built on generative models rather than fixed scripts. Criminal liability generally rests on a mental element—intent, knowledge or recklessness—and civil liability on foreseeability. Either way, culpability tracks what the defendant meant to do or should have seen coming. A generative agent given a broad goal can produce a specific, unlawful instruction that the person who deployed it never wrote, never reviewed, and could not have predicted. That gap between what a human authorized and what a machine actually did is not simply an old accountability problem wearing new clothes; it is a new one, because no earlier tool routinely generated instructions its own operator could not foresee.
This is not a hypothetical dressed up as a warning. It is the same “pay automatically once a condition is verified” architecture that scholars have worried about for years in writing about smart contracts — except now it is live, ordinary, and processing real tasks today.
Money still has to reach a bank
I checked the platform’s payment page myself. Settlement runs through ordinary card and bank-transfer rails. A marketplace that lets anyone, anywhere, rent out an hour of their physical presence is, in principle, an extraordinary opportunity for people who have few other ways to earn foreign income. But the conventional payment system is exactly what has historically excluded those same people: it requires a bank account able to receive international transfers, a card accepted by global processors, and a national banking system not cut off by sanctions or by the correspondent banks that quietly refuse to deal with it. The workers who would benefit most from a borderless, task-by-task labor market—people in sanctioned countries, in countries with thin banking infrastructure, in currencies international processors treat as too risky to touch—are the same workers most likely to be unable to collect payment from it at all. The platform does not need to build a discriminatory system to exclude them. The existing financial system already does that work for it.
I do not have to imagine this part. When I signed up for RentAHuman from Iran, the platform told me directly, during registration, that bounties were not available in my country. No specific law was cited, and none needed to be — a blanket nationality filter did the work instead, before I had posted or accepted a single task. This is a small, personal instance of a much larger pattern: compliance software tends to block far more than sanctions law actually requires. An Iranian graduate student can be denied an ordinary AI subscription or a chess website membership that no regulation actually forbids her from buying, for the same reason a task marketplace denied me its bounties: “Iranian” was coded into a system somewhere as a blanket risk flag, not weighed as a fact about a particular transaction. A labor marketplace screened by the same kind of automated compliance check is likely to exclude exactly the workers who need flexible, borderless income the most, on the basis of a nationality that no algorithm was ever asked to justify treating as disqualifying.
An old word for a new arrangement
The platform’s name is doing more work than its founders may have intended. “Rent” is an old word, and for most of human history it was not unusual to apply it to people rather than to things. Classical Islamic jurisprudence treated the hiring of persons as a form of ijara, or lease—the same contractual category used for a house or an animal. Some scholars working in this tradition have argued that this framing sits uneasily with human dignity, precisely because it treats a person’s labor as a leasable thing rather than treating the person as a rights-holding party to an agreement—a remnant of the era of slavery. The rise of modern statutory labor law effectively displaced this older contractual logic for exactly that reason, in Muslim-majority legal systems as much as anywhere else.
A parallel objection has come from an entirely different direction, with no debt to Islamic law at all: 19th-century American labor activists described selling one’s labor as selling oneself. “When you sell your product, you retain your person. But when you sell your labor, you sell yourself,” a labor newspaper, the Voice of Industry, wrote in 1848—not a fringe complaint, but a view historians have traced through mainstream free-labor ideology, including Lincoln’s own arguments against permanent wage status.
Two traditions with no shared history and no shared vocabulary arrived, independently, at the same discomfort: that renting out a person’s labor sits differently from renting out a person’s chair.
A platform that revives the word “rent” for human labor, whatever its founders intended by it, is reopening a legal question more than one legal tradition thought it had already closed.
Before the law defers to the algorithm
None of this is an argument to shut the idea down. A marketplace that lets a machine hire a human to do what the machine cannot is genuinely useful, and the demand for it is real. But being useful has never been the test for legal recognition. Dan-Cohen’s real point—buried in a book about mainframe computers, decades before anyone could build what he described — was that we should judge an arrangement by where control actually sits, not by where the paperwork says it sits. Before it hands an AI-agent labor market the recognition it eventually gave the gig economy, the law has to answer who the other party is, what duty it owes, and who is responsible when the instruction it gives should never have been given in the first place. The questions predate RentAHuman. The platform just made them impossible to ignore.
The author is an attorney licensed with the Iran Central Bar Association. He holds a master’s degree in International Commercial and Economic Law from the University of Tehran.
