Amazon will shut down Mechanical Turk on September 30, 2026, ending a 21-year-old crowdsourced work platform that Jeff Bezos once called “artificial artificial intelligence” and that played an outsized role in building the datasets behind modern AI.
The shutdown was announced in a customer email first reported by CNBC. The wording was blunt: “Following an assessment, Amazon made the decision to close AWS Mechanical Turk, effective September 30, 2026.” AWS had already stopped accepting new customer registrations on July 30, 2026, closing sister annotation services SageMaker Ground Truth and Amazon Augmented AI to new users on the same day. Existing customers now have roughly five weeks to migrate their workflows.
A Platform That Trained the AI That Replaced It
Mechanical Turk launched in 2005, named after the 18th-century chess-playing automaton secretly controlled by a human. The platform connected requesters with hundreds of thousands of workers, known as “Turkers,” who completed so-called Human Intelligence Tasks: audio transcription, image captioning, data labeling, and survey completion. Tasks paid pennies on average, with one worker estimating the typical payout at around 20 cents per task.
At its peak, the platform had over 500,000 registered workers. For years, it was the go-to source for the labeled data that machine learning researchers needed to train models. Bezos described it in 2007 as a service that farmed out tasks that were trivial for people but too hard for computers, a temporary solution until AI caught up.
The irony is that AI did catch up, and then it came for the workers themselves. A 2023 study from EPFL estimated that 33 to 46 percent of surveyed MTurk workers were completing text-based tasks using large language models like ChatGPT, meaning humans were being paid to relay machine outputs. The study found that the human-judgment premise underlying the platform had fundamentally cracked, as workers routed tasks through the very AI systems they helped train.
Scale AI, Mercor Fill the Vacuum
A new generation of AI-native data-labeling companies had already pulled requesters away from Mechanical Turk before Amazon pulled the plug. Scale AI, Mercor, and Prolific offer similar crowdsourced work with more modern tooling, faster turnaround, and quality controls designed for the LLM era. Mercor has been actively hunting for human signal data in recently shuttered startup Slack archives, seeking the kind of organic worker feedback that MTurk once produced.
The closure raises questions about where the remaining MTurk workload will land. If requesters land mostly with Scale AI and Mercor rather than AWS’s own SageMaker Ground Truth, it will signal that Amazon has ceded the data-labeling layer to third-party specialists. Amazon has not outlined a formal migration path from Mechanical Turk to any successor service.
“The platform that trained a generation of AI models is being killed by the models it helped enable,” AI Weekly noted in its coverage of the shutdown.
What Comes Next
The shutdown is part of a broader consolidation in AWS’s data services portfolio. SageMaker Ground Truth and Amazon Augmented AI, both related to human-in-the-loop data processing, are already closed to new signups. The move comes as the AI industry shifts from human annotation toward synthetic data generation, reinforcement learning from human feedback via specialized platforms, and automated evaluation pipelines that do not require crowdsourced workers.
For the workers who depended on MTurk as a source of flexible income, the closure marks the end of an era that never quite delivered on its promise. Many Turkers reported low pay, inconsistent work availability, and opaque rejection decisions from requesters. The platform’s decline was gradual, stretching over years as AI capabilities improved and alternatives emerged. The formal shutdown simply makes the end official.
Mechanical Turk’s legacy is complicated. It democratized access to micro-tasks and gave researchers a way to label data at scale when no automated alternative existed. But it also pioneered a model of precarious gig work that paid workers far less than minimum wage in many cases, with no benefits or protections. The platform that Bezos described as a stopgap until computers could do the work has now been rendered obsolete by the very progress it helped accelerate.
discussion