Amazon is shutting down Mechanical Turk after 21 years of operation, ending the human-work marketplace that played a foundational, if often under-recognised, role in training the machine-learning systems that underpin much of today's artificial intelligence industry, according to reporting published August 26, 2026.
Launched in 2005, Mechanical Turk allowed businesses and researchers to outsource small, discrete tasks — known as human intelligence tasks — to a global pool of independent workers, who were paid modest sums to complete work such as image labelling, data verification, transcription and content moderation. While the platform served many purposes over its two-decade existence, one of its most consequential legacies was the role it played in generating the vast, human-labelled datasets that trained early machine-learning and computer-vision systems, work that in many respects laid the groundwork for the deep learning revolution that would eventually give rise to today's large language models.
The closure of Mechanical Turk after more than two decades marks a symbolic bookend to a particular era of AI development, one built on the systematic, distributed human labour required to annotate and structure the raw data that machine-learning models needed to learn from. As AI systems have grown more sophisticated, the industry's data needs have evolved considerably, with modern large language models increasingly trained on vast quantities of existing internet text and multimodal content, supplemented by more specialised forms of human feedback such as reinforcement learning from human feedback, rather than relying primarily on the kind of discrete, task-based labelling work that defined Mechanical Turk's core offering.
Amazon's decision to wind down the platform reflects this broader shift in how the AI industry sources the human input still required to train and refine its systems, with many AI companies now employing more specialised, higher-skilled human feedback and evaluation workflows rather than the large-scale, low-cost crowdsourced labour model that Mechanical Turk pioneered. The rise of dedicated data-labelling and human-feedback companies serving AI developers directly has also likely reduced demand for Mechanical Turk's more generalised marketplace model.
For the broader gig economy and crowdsourced labour landscape, the closure removes one of the longest-running and most recognisable platforms in the category, one that despite persistent criticism over low pay rates and inconsistent work availability, nonetheless provided a meaningful income source for many workers globally, particularly in regions with limited access to other forms of flexible, internet-based earning opportunities.
The shutdown arrives during a week marked by a broader wave of consequential technology industry developments, including Meta's multi-billion-dollar settlement over claims related to teen platform design, OpenAI's claims regarding its custom inference chip outperforming Nvidia's leading hardware, and Samsung's unveiling of new memory technology aimed at accelerating AI inference — developments that collectively illustrate an AI industry whose infrastructure, talent and labour needs have evolved dramatically since the earlier era Mechanical Turk helped enable.
As Amazon closes this particular chapter of its technology portfolio, the platform's legacy is likely to be remembered less for its commercial scale within Amazon's broader business — Mechanical Turk was never a major revenue driver for the company — and more for the foundational, if largely invisible, role it played in supplying the human-generated training data that helped make the current generation of AI systems possible in the first place.
