Amazon's Mechanical Turk to Shut Down in September 2026

Amazon has announced that Mechanical Turk will permanently close on September 30, 2026, ending a service that has operated since 2005. The closure notice is now displayed across the service's official pages, including the homepage, FAQ page, Developer Resources page, and the worker How It Works page (mturk.com, mturk.com/help, mturk.com/resources, mturk.com/worker/how-it-works).

Amazon stated that the decision followed an assessment of its programs, tools, and services. No further detail on the rationale has been provided beyond that characterization. The company directed current Workers and Requesters to the service's FAQ page for guidance on preparing for the shutdown (mturk.com).

Mechanical Turk launched in 2005 as a crowdsourcing marketplace where human workers completed small tasks that were difficult for automated systems to handle at scale: image labeling, data validation, content moderation, sentiment classification, and similar micro-work. Requesters posted Human Intelligence Tasks, or HITs — individual units of work broken down into small, repeatable steps — which Workers could accept and complete for payment. Jeff Bezos once described the service as "artificial AI" (CNBC), a phrase that captured the essential premise: tasks presented as machine-completed were in fact routed to humans.

For the better part of two decades, Mechanical Turk occupied a specific and genuinely useful niche in the machine learning pipeline. Training data annotation, benchmark dataset construction, and human evaluation of model outputs all relied on the platform at various points. It was never the only option, but its API-driven task distribution model and pay-per-HIT pricing made it accessible to research teams and startups that lacked the scale to contract with dedicated data-labeling vendors. Academic researchers in particular used the platform heavily for behavioral studies, survey distribution, and experimental subject recruitment.

The service also drew sustained criticism over the years. Worker compensation was frequently low, often well below minimum wage when effective hourly rates were calculated. Task quality control mechanisms favored Requesters, who could reject completed work at their discretion. The platform's terms gave Workers limited recourse. These structural issues were documented in numerous academic studies and investigative reports across the platform's lifespan, though none of that backstory is new.

What is new is the closure itself, and the specific question it raises about what fills the gap for teams still relying on the platform. The September 30 deadline gives current users roughly a month to transition workflows, withdraw earnings, and identify alternatives. For data annotation, the market has matured considerably since 2005: dedicated labeling platforms, synthetic data generation, and model-assisted annotation pipelines have reduced dependence on raw crowdsourced labor for many use cases. But for ad hoc, low-volume human evaluation tasks, Mechanical Turk remained one of the more accessible options.

The FAQ page at mturk.com/help provides instructions for Workers on withdrawing remaining balances and for Requesters on managing outstanding tasks. The notice appears on all primary entry points to the service, leaving little ambiguity about the timeline.

The broader context here is that the closure of a 21-year-old service that Bezos himself framed as "artificial AI" lands at an interesting inflection point. The label was always half-joke, half-product-description. Human labor was the engine behind tasks that could not yet be automated. Two decades on, large language models can now handle many of the workloads that once flowed through Mechanical Turk: text classification, content moderation triage, sentiment scoring, even basic image description. The tasks that remain genuinely difficult for models are precisely the ones where crowdsourced, low-paid micro-labor was always the weakest solution.

The platforms that have grown up around more structured data annotation, with managed workforces, quality controls, and integration into ML training pipelines, reflect a market that moved beyond Mechanical Turk's original model some time ago. The closure formalizes that transition.

For Workers who depended on the platform for income, the timeline is short and the alternatives are not always obvious. For Requesters with active pipelines, the transition window demands immediate attention. The FAQ is the authoritative resource for next steps.