The Amazon Mechanical Turk closure process begins on 30 July 2026, when the platform stops accepting new customers, leaving a service that once shaped debates about crowdsourced labour quietly on life support. Amazon Web Services says the decision followed ‘careful consideration’ and that existing customers can continue as normal, but the company has no plans to introduce new features.
In other words: the lights stay on, but nobody is adding any new rooms.
Twenty Years of Humans Pretending to Be Machines
Amazon Web Services launched Mechanical Turk back in 2005 as a marketplace for microtasks: things a computer could not reliably do, like picking out the emotional tone in a sentence or ticking a CAPTCHA box. Workers were paid fractions of a dollar per task. The name was no accident. The original 18th-century ‘Mechanical Turk’ was a chess-playing automaton that turned out to have a human hidden inside, and the platform wore that irony proudly.
For a stretch, it sat at the centre of genuine ethical arguments about gig-economy labour. It also cropped up at the edges of the Facebook-Cambridge Analytica scandal, used to gather and process data during those early, chaotic years of social media research.
Then came the AI boom, and Mechanical Turk found a new pitch. At AWS re:Invent 2018, Amazon launched SageMaker Ground Truth, a data-labelling service that plugged directly into the Mechanical Turk workforce. The promise was that automated labelling could cut costs by up to 70 per cent, with human workers stepping in wherever the machine was not confident enough. The platform had reinvented itself as the unglamorous engine room of the neural-network era: less ‘click here to identify a bus’ and more ‘help us teach a self-driving car what a bus looks like.’ SageMaker Ground Truth even extended into labelling 3D point cloud data, with AWS describing a built-in editor that distributed annotation tasks automatically across private, third-party, or Mechanical Turk workers.
Amazon Mechanical Turk Closure and the AI Pivot That Made It Redundant
The real twist came when the tools Mechanical Turk helped train started turning up inside Mechanical Turk itself.
A preprint study submitted to arXiv on 13 June 2023 by Veniamin Veselovsky, Manoel Horta Ribeiro, and Robert West of EPFL, titled Artificial Artificial Artificial Intelligence: Crowd Workers Widely Use Large Language Models for Text Production Tasks, found that between 33% and 46% of Mechanical Turk workers on an abstract summarisation task were using large language models to complete their work. The researchers used keystroke detection and synthetic text classification to reach that estimate.
The implications cut two ways. First, if workers were feeding tasks to an LLM and submitting the output, the data being returned for AI training was itself generated by AI, creating a feedback loop that quietly corrupted the value of the whole exercise. Second, and more brutally for the platform’s business case: if a language model can do the task, why hire the human to relay it?
There was also a less flattering use that Mechanical Turk quietly enabled for years. A certain category of tech product, marketed as AI-powered, was in practice being handled by human workers on the platform. The original Turk hoax, running on modern infrastructure.
This week, after the closure announcement circulated, a Reddit user put it bluntly: the platform had died ‘years ago,’ abandoned by workers and researchers alike as bots and fraud eroded the quality of results. The same user predicted someone at Amazon would eventually decide keeping the servers running was not worth the cost and pull the plug entirely.
That prediction may already be in motion. Closing to new customers is the standard first step before a service is quietly retired: stop the inflow, let the existing base run down, then make the infrastructure decision. AWS is not saying that yet, but it is not saying the opposite either.
The platform that spent two decades teaching machines to think is now waiting to find out whether anyone still needs it to.
