Amazon to shut down Mechanical Turk, the platform that quietly trained the AI boom

Amazon will permanently close Mechanical Turk, the crowdsourced labor marketplace it launched in 2005, on September 30, 2026, ending one of the internet's oldest platforms for paid human digital labor just as the artificial intelligence systems it helped build have made much of that labor obsolete. A notice posted to the service's Mechanical Turk homepage tells workers and businesses the platform "will close September 30, 2026" and directs them to a help page to prepare for the shutdown.

The closure notice is terse. "Following an assessment, we've made the decision to close AWS Mechanical Turk, effective September 30, 2026," the site states, without elaborating on the reasoning. Amazon spokesperson Montana MacLachlan said the company remains focused on delivering the best possible experience for its Amazon and AWS customers, according to CNBC's report on the shutdown.

The numbers

Mechanical Turk, known to its users as MTurk, connected businesses and researchers with workers who completed small digital jobs called Human Intelligence Tasks, or HITs, for pay that often ran to a few cents apiece. At its peak the marketplace drew more than 500,000 registered workers, according to figures widely reported around the shutdown. Amazon had already begun winding the service down in stages: it stopped accepting new customers on July 30, 2026, a milestone confirmed on AWS's own service-availability update page, which lists Mechanical Turk among several SageMaker AI features moved to "maintenance" status ahead of full retirement. TechCrunch reported at the time that AWS said it would continue investing in security and availability for existing users but had no plans to add new features — language that in hindsight reads as a preview of Tuesday's closure notice.

From "artificial artificial intelligence" to real AI

Amazon founder Jeff Bezos coined the platform's tagline, calling it "artificial artificial intelligence" — a wink at the fact that the service used human workers to complete simple tasks, such as labeling images, transcribing audio and moderating content, that computers of the era could not reliably do themselves. The service's name and logo referenced the 18th-century chess-playing automaton "The Turk," a machine later revealed to have concealed a human operator inside its cabinet.

For most of the 2010s, MTurk was also a workhorse of academic research, used by social scientists and psychologists to recruit large, low-cost survey samples. By the late 2010s Amazon had repositioned it primarily as a labeling workforce for machine learning, feeding it directly into Amazon SageMaker Ground Truth and Amazon Augmented AI, the company's tools for annotating data used to train AI models. That integration is now unwinding along with the marketplace itself: AWS's documentation states plainly that "Amazon Mechanical Turk will permanently close on September 30, 2026," and instructs SageMaker customers who rely on the Mechanical Turk workforce option to migrate to a private or vendor-supplied workforce before the deadline.

The irony did not go unnoticed among researchers who study the platform. A 2023 academic survey found that as many as 46% of Mechanical Turk workers polled admitted to using large language models to help complete tasks — meaning a service built to supply verifiably human labor had, in its final years, become populated in part by the very AI systems it was meant to help train.

Who's affected

The shutdown lands hardest on two groups. The first is the "turkers" themselves — a workforce that at its height spanned well over 100 countries and included people who relied on the platform for supplemental income as well as those, often disabled or homebound, who depended on it as a primary source of earnings. The second is a smaller set of businesses, including insurance and travel companies, that Fast Company reported still use MTurk for routine data-processing work and must now find a replacement vendor before the end of September.

Amazon's exit leaves the field to a newer generation of AI-training data companies — among them Scale AI, Mercor and Prolific — that have built businesses specifically around recruiting and managing workers to label, rate and fine-tune outputs for large language models, a market that has grown substantially since generative AI systems took off. TheNextWeb noted that MTurk had been losing workers and business to these rivals for years, as Amazon appeared to slow investment in the platform's tools and anti-fraud systems even as competitors expanded.

Reaction

Krista Pawloski, a longtime Mechanical Turk worker and organizer with Turkopticon, the worker-advocacy group that has tracked conditions on the platform since 2008, said the closure was not a surprise to those who had watched the marketplace decline.

"MTurk had been in decline for years. Amazon put fewer resources into improving it while rivals arrived, and many turkers moved to competing platforms," Pawloski said, adding of the workers still active on the site: "There's some people that still pretty much still do it full time. They're concerned now."

Pawloski began working on Mechanical Turk in 2008 to supplement income during maternity leave and later turned to it as a primary source of income after losing a job while caring for a son with special needs, before becoming a Turkopticon organizer in 2019 — a trajectory she has said mirrors that of many long-tenured turkers now facing the platform's end.

What happens next

Amazon has given workers and business customers roughly five weeks from the closure notice to withdraw earnings, export data and migrate workflows before the September 30 deadline. The company has not said whether it will offer a successor product, and its public statements have so far framed the decision as the outcome of a routine internal review rather than a strategic pivot. For an industry watching AI labor markets consolidate around specialized data-training firms, Mechanical Turk's exit marks the end of the platform many credit with proving, at scale, that paid human micro-labor could be organized over the internet — the same insight that underpins the AI-training economy now replacing it.