Amazon Closes Mechanical Turk to New Customers After Two Decades

Amazon will stop accepting new customers for Mechanical Turk on July 30, 2026, closing new-account access to the crowdsourced task marketplace that has fed human-labeled data into machine learning pipelines for two decades. TechCrunch reported the change on July 5, and Amazon confirms it directly: a maintenance banner on mturk.com states plainly that the service “will be closed to new customers, effective July 30, 2026” and that “existing users will not be impacted by this change.”

The wording matters for anyone with a labeling pipeline that touches Mechanical Turk today. This is not a shutdown. It is a hard stop on new entrants to a service that helped define what “human-in-the-loop” meant for machine learning before that phrase was common vocabulary.

What Amazon is actually changing

Amazon Web Services says the decision followed “careful consideration.” Its own language, repeated verbatim on both the public mturk.com banner and in Amazon SageMaker AI’s developer documentation, is specific: “Existing customers can continue to use the service as normal. AWS continues to invest in security and availability improvements for Mechanical Turk, but we do not plan to introduce new features.”

That SageMaker documentation page is worth reading on its own, because it shows the change reaches further than the standalone requester portal most people associate with Mechanical Turk. The same notice appears on the page that explains how to route SageMaker Ground Truth labeling jobs and Amazon Augmented AI (A2I) human review loops to the Mechanical Turk workforce through a shared “public-crowd” work team. In other words, July 30 closes new-customer access to Mechanical Turk both as a standalone marketplace and as a workforce option inside SageMaker’s labeling tools. Teams already running jobs against that work team keep running; teams that have never used it cannot start after that date.

Two decades from chess hoax to data-labeling pipeline

Mechanical Turk takes its name from an 18th-century chess-playing automaton, “The Turk,” built for the Habsburg court and later exhibited across Europe and the United States. The machine appeared to play chess autonomously, but a human chess player was concealed inside the cabinet, operating it from within. Amazon launched its namesake service in 2005 as a marketplace for small, paid “Human Intelligence Tasks,” or HITs, that resisted full automation at the time: solving CAPTCHAs, tagging the sentiment of a sentence, and similar small tasks.

The service’s role changed as machine learning matured. Beginning in 2018, Amazon positioned Mechanical Turk as a way for companies to annotate the training data behind neural networks, and it became one of the workforce options built into SageMaker Ground Truth. According to TechCrunch’s reporting, a 2023 analysis found that between 33% and 46% of workers on the platform were using large language models to complete tasks: a detail worth flagging for anyone relying on older Mechanical Turk-labeled datasets, since it suggests some “human” ground truth from that period was already partly machine-generated before today’s model generation existed.

Part of a wider SageMaker cleanup

AWS’s own Lifecycle documentation, under “Services in Sunset,” logs the Mechanical Turk change with an announcement date of June 30, 2026, alongside eight other SageMaker AI feature changes announced the same day: Amazon A2I, Clarify, Debugger, GeoSpatial, Ground Truth, Model Monitor, Role Manager, and Studio Lab. That grouping places Mechanical Turk’s closure to new customers inside a broader consolidation of older SageMaker tooling, rather than a standalone call about crowdsourced labor specifically.

Why it matters for AI teams

The practical planning window is narrow. Any team that expected to reach for the Mechanical Turk workforce for the first time (whether through the standalone requester console or a new SageMaker Ground Truth labeling job) needs an alternative lined up before July 30, 2026. Ground Truth’s other workforce options (a private, self-managed labeler pool or a third-party vendor workforce) are not mentioned in either notice and are unaffected; the closure is scoped specifically to the public Mechanical Turk crowd.

For teams already running Mechanical Turk-backed pipelines, nothing breaks on July 30. The more useful exercise is auditing where existing labeled datasets came from and when, given the documented rise of LLM-assisted work on the platform. A dataset labeled through Mechanical Turk in 2023 or later may carry more machine influence than its provenance records suggest.

Bottom line

Mechanical Turk is not disappearing on July 30, 2026, but its era of open enrollment is. The service that gave “human-in-the-loop” its first large-scale, commercial form is now closed to newcomers, folded into a broader cleanup of SageMaker’s older feature set, at the same moment the AI industry is still arguing over how much of that “human” input was human at all.

Sources: TechCrunch report on Amazon’s Mechanical Turk decision, Amazon Mechanical Turk’s public maintenance notice, and Amazon SageMaker AI documentation on the Mechanical Turk workforce.

Featured image: 1789 engraving by Joseph Racknitz depicting Wolfgang von Kempelen’s “Turk” chess-playing automaton, the hoax the Mechanical Turk service is named after, seated at its cabinet with the doors open to show the machinery inside. Via Wikimedia Commons / Humboldt University Library, public domain (PD-Art/PD-old-100); cropped, resized, and converted to WebP.