Quick Answer: Amazon announced on August 25, 2026, that it will permanently shut down Amazon Mechanical Turk (MTurk) on September 30, 2026, ending a 21-year-old crowdsourced labor platform that once connected more than 500,000 workers with businesses needing human judgment for tasks computers couldn’t handle. Launched in 2005 and famously described by Jeff Bezos as “artificial artificial intelligence,” MTurk paid workers a few cents at a time to label data, transcribe audio, and complete surveys, work that helped train the very machine-learning models that eventually made much of that labor unnecessary. Amazon is also shutting down two related services, SageMaker Ground Truth and Amazon Augmented AI, on the same date, effectively exiting the human-data infrastructure business entirely.
What’s Actually Happening
On Tuesday, August 25, 2026, Amazon posted a notice on the Mechanical Turk website announcing the platform’s permanent closure. “We regularly evaluate our programs, tools and services and make adjustments based on those assessments,” the notice read. “Following an assessment, we’ve made the decision to close AWS Mechanical Turk, effective September 30, 2026.” Beyond that banner, Amazon has published no separate statement explaining the decision in detail, and the company has not commented further on why now, specifically, after more than two decades of operation.
The closure wasn’t entirely unannounced. Amazon stopped accepting new Mechanical Turk customers on July 30, 2026, alongside two related services, SageMaker Ground Truth and Amazon Augmented AI, a move many longtime workers and researchers interpreted at the time as a clear signal that a full shutdown was coming. That signal turned out to be correct.
What Was Mechanical Turk?
Amazon Mechanical Turk, launched in 2005, created a marketplace where companies and researchers could pay people to complete small online jobs formally known as Human Intelligence Tasks, or HITs. These were simple, repetitive assignments that, at the time, remained genuinely difficult for computers to do reliably: labeling images, transcribing audio and video, moderating content, validating data, and completing surveys. Workers, commonly called “Turkers,” were typically paid just a few cents per task. Amazon originally built the system internally to help organize and label the enormous volume of data flowing through its own online marketplace, according to reporting from The New York Times, before opening the tool up to outside businesses and academic researchers.
At its peak, the platform served more than 500,000 workers worldwide, and it became something close to standard infrastructure for a generation of academic behavioral research, survey studies, and, especially in more recent years, the human-labeled datasets that early machine learning models were trained on.
Where the Name Comes From
The service’s name is a direct reference to the original Mechanical Turk, an 18th-century device built to look like a chess-playing automaton that could defeat human opponents. In reality, the machine concealed a human chess master hidden inside its cabinet, secretly operating the mechanism and making every move. It was, in other words, an elaborate illusion of artificial intelligence powered entirely by a hidden person. Amazon’s crowdsourcing platform borrowed that name deliberately: the tasks routed through MTurk were often marketed to businesses as if they might be handled by software, when in fact a real person on the other end was doing the thinking. Jeff Bezos captured that irony directly, describing the service as “artificial artificial intelligence,” a phrase that stuck for the better part of two decades.
Timeline: From Launch to Shutdown
- 2005: Amazon launches Mechanical Turk, initially to help label data across its own retail marketplace.
- Peak years: The platform grows to serve more than 500,000 workers globally, becoming widely used for academic research and early AI training data.
- 2023: A study by researchers in Switzerland (EPFL) finds that 33 to 46 percent of surveyed MTurk workers may have used AI tools to complete tasks, undermining the platform’s core premise of purely human-generated work.
- July 30, 2026: Amazon stops accepting new customers for Mechanical Turk, SageMaker Ground Truth, and Amazon Augmented AI.
- August 25, 2026: Amazon formally announces MTurk will permanently close on September 30, 2026.
- September 30, 2026: Mechanical Turk, SageMaker Ground Truth, and Amazon Augmented AI are all scheduled to shut down for good.
Why Amazon Is Shutting It Down Now
Krista Pawloski, a longtime data worker and organizer with Turkopticon, a worker advocacy group focused on MTurk, told CNBC the platform had been “in decline” for some time, with Amazon appearing to invest fewer resources into improving it as competing data-labeling services emerged and gradually drew workers away. That decline tracks with a broader shift in how AI companies source training and evaluation data. A newer generation of data-labeling startups, including Scale AI, Mercor, and Prolific, has emerged specifically to recruit and manage workers for AI training pipelines, often with more specialized tooling, quality controls, and pay structures than MTurk’s open, largely unmoderated marketplace ever offered.
Amazon’s own explanation stays deliberately vague, framing the closure as the outcome of a routine internal assessment rather than pointing to any single cause. But the practical result is the same either way: a platform that helped define an entire category of online work, crowdsourced “microtasks,” is being wound down just as the AI industry it helped build has moved on to more specialized alternatives.
The Irony: AI Workers Using AI
Perhaps the most striking detail in MTurk’s final chapter is how directly it undercuts the platform’s original premise. Mechanical Turk existed because computers, in 2005, simply couldn’t do certain kinds of judgment-based work. By 2023, that premise had started to collapse from the inside: a study conducted by researchers at EPFL in Switzerland found that somewhere between 33 and 46 percent of MTurk workers participating in one experiment appeared to be using large language models to complete their assigned tasks. In effect, a platform built specifically to supply “real” human intelligence as an alternative to artificial intelligence had begun quietly filling itself with AI-assisted output instead, blurring the line between human and machine-generated work in exactly the space that line was supposed to matter most.
That shift matters enormously for anyone relying on MTurk data for research or model training, since the entire value proposition depended on the labor being genuinely human. Once that assumption became unreliable, much of the platform’s usefulness to serious AI labs and academic researchers eroded along with it, even before Amazon’s own resourcing decisions accelerated the decline.
What’s Replacing It: The New Data-Labeling Economy
MTurk’s decline lines up almost exactly with the rise of a new category of AI-focused labor platforms built specifically around the needs of modern model training and evaluation. Scale AI, Mercor, and Prolific have all positioned themselves as more specialized alternatives, offering better-vetted worker pools, task types built around reinforcement learning from human feedback (RLHF), model evaluation, and red-teaming, and pay structures that go well beyond MTurk’s traditional few-cents-per-task model. Where MTurk operated as an open, largely self-service marketplace that anyone could post simple tasks to, these newer platforms tend to run more managed, curated pipelines built explicitly for the demands of frontier AI labs, a fundamentally different business than the general-purpose crowdsourcing marketplace MTurk represented for most of its life.
Amazon’s Broader AI Strategy This Week
The Mechanical Turk closure didn’t happen in isolation. It landed in the same news cycle as several other major AI infrastructure stories, underscoring just how fast the underlying economics of building and training AI systems are shifting in 2026. The same week, reports emerged of a major AI hub acquisition in the open-source model space, a reminder that consolidation is happening across nearly every layer of the AI stack simultaneously, from the chips that run models, to the platforms that distribute them, to the human labor that once trained them. Amazon’s decision to exit human-data labeling infrastructure altogether, rather than simply modernizing MTurk into something more competitive with Scale AI or Mercor, suggests the company has concluded that specialized, narrowly focused competitors have simply out-executed its original general-purpose approach in this particular corner of the AI supply chain.
What Happens to Existing Workers and Requesters
Amazon has given workers and requesters roughly five weeks between the closure announcement and the actual shutdown date to wind down their activity on the platform. According to the official closure FAQ, the single most important action for workers is to verify their payment settings immediately, confirming a valid payment method and transfer schedule in their MTurk Worker account, since unresolved balances need to be settled before the platform disappears. Requesters, the businesses and researchers who post tasks, are expected to receive full refunds of any remaining account balance within 30 days of the closure. Both groups are being directed to the platform’s FAQ page for further guidance on retrieving data and closing out active work before the September 30 deadline.
SageMaker Ground Truth and Amazon Augmented AI Also Shutting Down
A detail that received less attention than the MTurk closure itself is that Amazon is simultaneously shutting down SageMaker Ground Truth and Amazon Augmented AI (A2I) on the exact same date. SageMaker Ground Truth had been positioned by Amazon as the more managed, enterprise-friendly successor to MTurk’s open-crowd labeling model, while Amazon Augmented AI handled human-review workflows for machine learning predictions that needed a person to double-check them. The official closure FAQ confirms that the “Amazon Mechanical Turk Worker” type will no longer be available for creating SageMaker labeling jobs or A2I human review workflows once the September 30 deadline passes. Taken together, these three closures mean Amazon isn’t simply retiring one aging product from an otherwise intact AWS portfolio, it’s exiting the human-data infrastructure business as a whole, at least under its own direct branding.
How This Fits the Bigger AI Infrastructure Shift
MTurk’s shutdown lands in the middle of a broader realignment of how the AI industry sources both the human labor and the raw compute that training and running modern models require. The same week Amazon confirmed this closure, OpenAI was busy publishing benchmark results for its own custom silicon, a reminder of how much of the current AI buildout is now about specialized infrastructure rather than general-purpose tools. Our coverage of OpenAI’s Jalapeño chip beating Nvidia on inference efficiency covers a similar dynamic on the hardware side: purpose-built, specialized tools increasingly displacing older, more general-purpose infrastructure as the industry matures.
A similar logic applies to open-source AI distribution. Our look at Nvidia’s reported $12.9 billion acquisition of Hugging Face covers how even the platforms where AI models themselves get shared and distributed are consolidating around a handful of specialized, well-resourced players rather than staying general-purpose and open, much the way data labeling has consolidated around companies like Scale AI and Mercor instead of Amazon’s original open marketplace.
What This Means for AI Research Going Forward
For academic researchers, MTurk’s closure removes a tool that had become deeply embedded in behavioral science, psychology, and social science research methodology over nearly two decades, well beyond its role in AI training data. Many published studies relied specifically on MTurk’s large, relatively low-cost, and demographically broad worker pool, and researchers will now need to migrate to alternative platforms such as Prolific, which has increasingly positioned itself as an academic-research-friendly option, or to newer AI-focused labor platforms that weren’t originally designed with survey research in mind. That transition isn’t necessarily seamless, since different platforms carry different worker demographics, pricing structures, and data quality characteristics that could complicate efforts to replicate or extend research originally conducted on MTurk.
MTurk’s Outsized Role in Early AI Research
It’s difficult to overstate how central Mechanical Turk became to the field of machine learning in its first fifteen or so years of existence, well before today’s frontier AI labs built their own specialized data pipelines. Foundational datasets used to train and benchmark early computer vision and natural language processing systems were frequently built, at least in part, using MTurk labor: workers drawing bounding boxes around objects in images, labeling sentiment in text snippets, or verifying whether two pieces of text meant the same thing. That work was slow, occasionally inconsistent, and famously underpaid relative to its economic value, but it was also genuinely foundational. Entire subfields of AI research effectively couldn’t have progressed at the pace they did without a cheap, scalable, on-demand supply of human judgment, and for most of the 2010s, MTurk was where that supply lived.
That history makes the platform’s closure feel less like the retirement of a minor Amazon side project and more like the quiet decommissioning of a piece of AI infrastructure that, in a very real sense, helped make the current AI boom possible in the first place, only to be rendered obsolete by the very technology it helped train.
What Happens Next
With the September 30, 2026 deadline just over a month away as of this writing, the immediate priority for anyone still active on the platform is straightforward: workers should confirm payment details are current, and requesters should retrieve any data or results they still need before access disappears. Beyond the individual wind-down, the more interesting question is where Amazon’s presence in AI training infrastructure goes from here. Exiting MTurk, SageMaker Ground Truth, and Amazon Augmented AI all at once suggests less a full retreat from AI infrastructure and more implicit acknowledgment that the human-data labeling business has moved decisively toward the specialized, purpose-built platforms that have emerged over the past few years, a market Amazon is stepping back from rather than trying to compete in directly.
Criticism MTurk Faced Over the Years
MTurk’s long run wasn’t without sustained criticism, much of it centered on worker pay and working conditions. Because tasks were typically priced at just a few cents each, effective hourly earnings for many Turkers came in well below minimum wage in the countries where most workers were based, a problem labor advocates and academics raised repeatedly over the platform’s lifetime. Turkopticon, the advocacy group whose organizer Krista Pawloski commented on the platform’s decline, was founded specifically to help workers rate and avoid exploitative requesters, filling a gap in worker protections that Amazon’s own platform never fully addressed. Researchers who relied on MTurk for academic studies also faced periodic scrutiny over whether the platform’s demographics, skewing toward workers with reliable internet access and a certain baseline of English fluency, produced genuinely representative samples for social science research, a methodological question that shaped how carefully some fields treated MTurk-sourced data even at the height of its popularity.
Frequently Asked Questions
When is Amazon Mechanical Turk shutting down?
Amazon Mechanical Turk will permanently close on September 30, 2026, roughly five weeks after the closure was announced on August 25, 2026.
Why did Amazon shut down Mechanical Turk?
Amazon described the closure as the result of a routine internal assessment. Reporting suggests the platform had been in decline for years as competing AI-focused data-labeling services drew workers and business away.
What was Amazon Mechanical Turk used for?
MTurk let businesses and researchers pay workers to complete small online jobs called Human Intelligence Tasks, including data labeling, audio and video transcription, content moderation, and survey completion.
Why was it called “Mechanical Turk”?
The name references an 18th-century chess-playing automaton that appeared to be artificial intelligence but actually concealed a hidden human chess master. Jeff Bezos used the phrase “artificial artificial intelligence” to describe Amazon’s platform in the same spirit.
What is replacing Mechanical Turk?
Newer AI-focused data-labeling platforms including Scale AI, Mercor, and Prolific have emerged to handle the kind of work MTurk once provided, often with more specialized tooling built for AI model training and evaluation.
Are other Amazon services closing at the same time?
Yes. SageMaker Ground Truth and Amazon Augmented AI are both scheduled to close on the same September 30, 2026 date, meaning Amazon is exiting human-data labeling infrastructure entirely.
What should current MTurk workers do?
Workers should immediately verify their payment method and transfer schedule in their MTurk account to ensure any remaining balance is properly settled before the platform closes.
What happens to requesters’ account balances?
Requesters are expected to receive full refunds of any remaining account balance within 30 days of the platform’s closure.
