Amazon has officially notified users of its more than two-decade-old crowdsourced work platform, Mechanical Turk, that the service will close on September 30, 2026. According to a customer email reported by CNBC, the company stated that 'Following an assessment, Amazon made the decision to close AWS Mechanical Turk, effective September 30, 2026'. The notice emphasized that 'We regularly evaluate our programs, tools, and services and make adjustments based on those assessments'. As reported by CNBC, Amazon had already stopped accepting new Mechanical Turk customers on July 30, 2026, alongside the closure of sister services AWS SageMaker Ground Truth and Amazon Augmented AI to new users on the same day. Existing customers now have approximately five weeks before the cutoff date.
Mechanical Turk connects large numbers of workers with basic online tasks known as 'Human Intelligence Tasks', which generally pay just a few cents each. As reported by Amazon, these tasks can involve data labelling, audio or video transcription, and completing surveys. The platform enables companies to harness collective intelligence from a global workforce for business process streamlining, data collection and analysis, and machine learning development. According to the company's website, workers are available 24 hours a day, 7 days a week and typically provide the fastest turnaround for human review tasks. However, Amazon's historical claim of more than 500,000 workers does not indicate current active participation before the closure. Workers typically earned around 20 cents per task, as reported by CNBC, with one aggrieved worker providing this estimate.
The platform's fundamental premise has been undermined by the very AI models it helped train. According to a 2023 EPFL study reported by TechCrunch, 33 to 46 percent of surveyed workers were completing text tasks using LLMs like ChatGPT, meaning humans were being paid to relay a machine's guesses. This represents a significant shift from the original concept of 'artificial artificial intelligence' that founder Jeff Bezos once described, where the service outsourced tasks that humans could perform easily but computers found difficult. The service farmed out tasks like audio transcription, image captioning and data labeling that were trivial for people but too hard for computers, but the models MTurk had helped train have now become good enough to complete these tasks themselves. As reported by CNBC, Amazon also looked to capitalize on AI training demand by marketing MTurk as a data annotation source for its SageMaker service.
The closure comes as a newer crop of AI-native data-labeling companies, including Scale AI, Mercor and Prolific, have already pulled requesters away from MTurk. According to CNBC, Mercor was recently seen hunting shuttered-startup Slack archives for the same kind of human signal MTurk used to produce. Insurance and travel companies that still depend on the service are seeking alternative platforms, with specialist providers such as Scale AI, Mercor, and Prolific becoming established alternatives for data annotation and model-training work. The transition exposes quality problems as AI models themselves advanced rapidly, prompting many workers to move elsewhere. Amazon has directed existing customers to a FAQ page to prepare for the closure, providing guidance on alternative platforms and transition strategies. As reported by Turkopticon organizer Krista Pawloski, 'workers are available 24 hours a day, 7 days a week' and typically provide the fastest turnaround for human review tasks.
According to The New York Times, the service takes its name from the original Mechanical Turk, an 18th-century chess-playing machine operated secretly by a human chess master. The tech giant reportedly developed the service to help handle vast amounts of data generated by its sprawling online store. The platform's closure represents the end of a service that has been integral to Amazon's data processing and AI development efforts over the past two decades. For workers, the loss represents a channel for remote piecework rather than just a change in where training data is sourced, with some people still doing the work full-time and concerned about the shutdown. As reported by Turkopticon organizer Krista Pawloski, 'there's some people that still pretty much still do it full time' and are now concerned about the closure. The service provided meaningful income and flexible hours with work that can be completed on a smartphone, similar to other gig jobs like ride-hail services or Amazon Flex package delivery.
