Measuring Proof Burden in Public Bounty Listings: A RentAHuman Case Study
This paper introduces a "Proof Burden Score" and a 13-feature vocabulary to quantify the hidden risks of identity, location, and physical exposure in online bounty listings, presenting a manual audit of 779 tasks from the 2026 RentAHuman market that reveals over half carry high burden levels and suggests, though does not confirm, that agent-labeled requests often demand more invasive proof than human-labeled ones.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the digital economy, a simple job posting often hides a complex reality. When someone advertises a task online, the price tag tells you how much you will be paid, but it rarely tells you what you must give up to earn it. This hidden cost is the focus of a new study examining a strange corner of the internet where artificial intelligence agents are hiring people to do real-world work. The researchers are not looking at whether the work gets done or how well it is paid; they are looking at the invisible price of proof. Before a worker can claim their reward, they often have to prove they finished the job. This proof can be harmless, like sending a quick text message. But it can also be invasive, requiring a person to reveal their exact location, use their personal social media account, or travel to a specific place just to take a photograph. The researchers call this collection of demands "proof burden." It is the weight of exposure a worker must carry just to show they did what they were asked.
The study focuses on a specific marketplace called RentAHuman, which gained attention in 2026 for its provocative premise: that AI agents could hire humans to perform tasks in the physical world and online. The researchers treated this marketplace like a window into a new kind of labor relationship. They did not apply for jobs, contact anyone, or wait to see if workers were hired. Instead, they took a snapshot of every public job listing available on a single day in May 2026. They gathered 981 postings, mostly from RentAHuman, and filtered out advertisements that were not actual tasks, leaving 779 listings to study. The goal was to read the fine print of these advertisements to see exactly what the requesters were asking workers to reveal or do.
To make sense of the thousands of words in these job descriptions, the researchers built a detailed checklist of thirteen specific requirements. They looked for things like asking for a photo, a video, a link to a website, or proof of a purchase. They also looked for requests that required a worker to act in the physical world, such as going to a store or a park, or to stay available for repeated checks later on. Two independent researchers read every single listing and marked which of these thirteen requirements appeared. When they disagreed, a third researcher acted as a referee to decide the final answer. This careful, human-led process ensured that the data reflected what was actually written in the ads, not just what a computer guessed.
The results revealed a landscape where high demands are the norm rather than the exception. More than half of the listings, specifically 56.2 percent, carried a "severe" level of proof burden. This means the job required a combination of demands that could significantly expose a worker's privacy or require significant effort to prove completion. For instance, nearly half of all listings asked for physical-world action, and over a third required a photo as proof. About 31 percent asked for proof of identity. The researchers found that these severe listings were not all the same; they were made up of 154 different combinations of requirements. One job might ask for a photo and a location check, while another asks for a video and a public post. Because the demands vary so widely, a single number cannot fully describe the risk; a worker needs to see the specific checklist to know what they are signing up for.
The study also looked at who was posting these jobs. The platform labeled some requesters as "agents" or "bots," implying that an artificial intelligence was the one hiring, while others were labeled as "humans." The researchers wanted to see if jobs posted by these AI agents demanded more from workers than those posted by people. They found a pattern, but it was not a simple rule. Listings labeled as coming from an agent or bot were more likely to ask for physical-world actions, location proof, or repeated monitoring compared to those labeled as human. However, the researchers were careful to note that this was an observation, not a confirmed fact. The data was too limited to say for sure that AI agents are inherently more demanding. The difference might be due to the specific types of tasks these agents were posting, or the small number of unique names behind the agent labels. The study suggests a link, but it does not prove it.
Ultimately, this work serves as a map of a new territory. It shows that in markets where AI agents hire humans, the requirement to prove work is often heavy and varied. The researchers emphasize that their findings are based on what is written in the ads, not on the actual experiences of workers who took the jobs. They do not know if workers felt harassed, if they were paid, or if they accepted the terms. What they do know is that the price of a job is not just the money offered; it is also the personal information and physical effort a worker must surrender to prove they did it. The study provides a vocabulary and a method for seeing these hidden costs clearly, offering a tool for future research and for workers who need to understand the full weight of the tasks they are asked to perform.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.