MIT study: RentAHuman bounties demand identity exposure from 56% of workers

A new MIT-led study, accepted at ACM HCOMP 2026, exposes the hidden costs of human-bounty work on AI-agent marketplaces. Researcher Iman YeckehZaare (MIT Center for Collective Intelligence and Honor Education) manually audited 981 listings from RentAHuman and Human Pages, narrowing to 779 legitimate bounty/task posts. Using two independent coders plus a blinded adjudicator, the team built a 13-requirement vocabulary covering evidence types, recurring monitoring, and physical-world action, then compressed it into a 0–5 Proof Burden Score. The results: 56.2% of listings scored 4 or 5, meaning workers are regularly asked to expose themselves—revealing identity or location, using personal accounts, posting publicly, acting in the physical world, or providing repeated evidence at later checks.

The paper also flags a striking pattern: agent-or-bot-labeled requester accounts included physical-world action, location proof, or recurring monitoring in 75.0% of listings, versus 55.3% for human-labeled accounts. However, the authors caution this comparison was post hoc, drawn from only 20 displayed names, and the labels are self-reported or platform-assigned—so it's a hypothesis, not a confirmed difference. The study's core contribution is descriptive: a vocabulary and scoring rubric that lets workers see what a listing truly entails before accepting—a checklist, not a single number. While RentAHuman's model of AI agents hiring humans could scale task delegation, this work highlights that "proof" burdens transfer risk and exposure onto human gig workers, often without transparent pricing. The authors note no worker-validated measure or automated detector exists yet, leaving that for future work.