--- title: "Amazon Shuts a 21 Year Old Gig Work Platform on September…" canonical: "https://www.metaintro.com/blog/amazon-mechanical-turk-shutdown-gig-workers-2026" language: "en" author: "drashtigarach" published: "2026-08-26T12:22:50.000Z" modified: "2026-10-02T19:45:46.354Z" --- [Back to Blog](/blog) [Gig Economy](/blog/tag/gig-economy)[AI](/blog/tag/ai)[Jobs](/blog/tag/jobs)[Technology](/blog/tag/technology)[News](/blog/tag/news) # Amazon Shuts a 21 Year Old Gig Work Platform on September 30 and What Its Workers Do Next Amazon closes Mechanical Turk on September 30 after 21 years, ending a platform that once served 500,000 workers. What data workers should do next. [![Drashti Garach](https://cdn.metaintro.com/rs:fill:40:40/q:72/plain/images/5719d740-e510-42bc-8017-e040d145f35f_1766029465094.png)Drashti Garach @DrashtiGarach](/blog/author/drashtigarach) [August 26, 2026](/blog/archive/2026/08)12 min read ![Amazon Shuts a 21 Year Old Gig Work Platform on September 30 and What Its Workers Do Next](https://cdn.metaintro.com/rs:fill:1200:675/q:78/plain/images/hero-amazon-mechanical-turk-shutdown-gig-workers-2026.png) [https://x.com/intent/tweet?text=Amazon%20Shuts%20a%2021%20Year%20Old%20Gig%20Work%20Platform%20on%20September%2030%20and%20What%20Its%20Workers%20Do%20Next&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026](https://x.com/intent/tweet?text=Amazon%20Shuts%20a%2021%20Year%20Old%20Gig%20Work%20Platform%20on%20September%2030%20and%20What%20Its%20Workers%20Do%20Next&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026)[http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026](http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026)[https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026&title=Amazon%20Shuts%20a%2021%20Year%20Old%20Gig%20Work%20Platform%20on%20September%2030%20and%20What%20Its%20Workers%20Do%20Next](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026&title=Amazon%20Shuts%20a%2021%20Year%20Old%20Gig%20Work%20Platform%20on%20September%2030%20and%20What%20Its%20Workers%20Do%20Next)[mailto:?subject=Amazon%20Shuts%20a%2021%20Year%20Old%20Gig%20Work%20Platform%20on%20September%2030%20and%20What%20Its%20Workers%20Do%20Next&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026](mailto:?subject=Amazon%20Shuts%20a%2021%20Year%20Old%20Gig%20Work%20Platform%20on%20September%2030%20and%20What%20Its%20Workers%20Do%20Next&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026) According to [CNBC](https://www.cnbc.com/2026/08/25/amazon-service-that-jeff-bezos-called-artificial-ai-is-shutting-down.html), [Amazon](https://www.amazon.com/) is shutting down [Mechanical Turk](https://www.mturk.com/) on September 30, 2026, closing a 21-year-old platform that matched workers with small digital jobs paying a few cents each. The notice posted to the site said the company had "made the decision to close AWS Mechanical Turk, effective September 30, 2026". For the people who earned money there, the practical question is where that income goes now. At [Metaintro](https://www.metaintro.com), we track how automation reshapes real jobs rather than how it looks in a press release, and this closure is a clear case of a work category being restructured rather than simply erased. The demand for human input into AI systems is still growing. The platform that made it famous is the part going away. ## What is Amazon closing on September 30? Mechanical Turk launched in 2005 as a way to farm out work that people find easy but computers find hard. [Amazon Web Services](https://aws.amazon.com/) ran it as a marketplace for what the platform called Human Intelligence Tasks, which covered labeling data, transcribing audio or video, and answering surveys. Each task typically paid out a few cents. Amazon founder Jeff Bezos described the service as "artificial artificial intelligence", because it quietly routed work to humans that software could not yet handle. The name came from an 18th century chess playing machine that appeared automated but concealed a human chess master inside. The scale was real. The platform at one point served more than 500,000 workers, known as turkers, according to Amazon's own figures. Amazon originally built it to help label data across its retail store. In its closing notice the company said only that it "regularly evaluate our programs, tools and services and make adjustments based on those assessments". A month before the announcement, Amazon stopped accepting new Mechanical Turk customers, which many workers read as a signal that the end was coming. ## Why did a platform with 500,000 workers wind down? The short answer is that the market moved and Amazon did not move with it. AI models advanced quickly after Mechanical Turk launched, and a crop of specialist data companies including [Scale AI](https://scale.com/) and [Prolific](https://www.prolific.com/) entered the market to recruit workers who train models. Amazon tried to capture some of that demand by marketing the platform as a data annotation source for [Amazon SageMaker](https://aws.amazon.com/sagemaker/), promoting that "Workers are available 24 hours a day, 7 days a week". That was not enough. Krista Pawloski, a data worker and organizer with the advocacy group [Turkopticon](https://turkopticon.net/), said the platform had been "in decline" in recent years and that Amazon appeared to invest fewer resources into improving it as rival services arrived, pushing many turkers onto competing platforms. There was also a quality problem that cut to the heart of what buyers were paying for. Researchers at [EPFL](https://www.epfl.ch/en/) published [a 2023 study](https://arxiv.org/abs/2306.07899) that replicated a summarization task on the platform and estimated, using keystroke detection and synthetic text classification, that 33 to 46 percent of crowd workers used large language models to complete it. When a client buys human judgment and receives machine output, the core product stops working. This is the same tension covered in Metaintro's reporting on [AI text watermarks in job applications](https://www.metaintro.com/blog/ai-text-watermarks-job-applications-2026), where the value of a human signal collapses once it becomes cheap to fake. ## What did the work actually pay? The pay was low enough that its disappearance is a smaller financial shock than the worker count suggests, and that is its own indictment. A study of the platform analyzed 2,676 workers completing 3.8 million tasks and found a median hourly wage of about 2 dollars, with only 4 percent of workers earning more than 7.25 dollars an hour. The gap was not because clients were cheap across the board. The average requester paid more than 11 dollars an hour, but the lower paying requesters posted far more work, so the typical worker spent most of their time on the worst paid listings. The [International Labour Organization](https://www.ilo.org/) found the same pattern across the wider crowdwork market. Its [survey of 3,500 workers in 75 countries](https://www.ilo.org/publications/digital-labour-platforms-and-future-work-towards-decent-work-online-world) working on five English speaking microtask platforms found average earnings of 4.43 dollars an hour when only paid work was counted, falling to 3.31 dollars once unpaid hours were included. Median earnings were 2.16 dollars an hour on that fuller measure. Nearly two thirds of American workers surveyed on Mechanical Turk earned less than the federal minimum wage of 7.25 dollars an hour, a threshold set by the [US Department of Labor](https://www.dol.gov/general/topic/wages/minimumwage). Workers averaged 24.5 hours a week on the platforms, of which 18.6 were paid and 6.2 were not. ## How does crowdwork pay differ around the world? Geography decided a great deal. The ILO survey found workers in Northern America earning 4.70 dollars an hour and workers in Europe and Central Asia earning 3.00 dollars, while earnings elsewhere ranged from 1.33 dollars in Africa to 2.22 dollars in Asia and the Pacific per hour of paid and unpaid work. The same task, routed through the same interface, paid a worker in one region roughly three times what it paid a worker in another. For readers in India and across Asia, that gap is the central fact about remote task work, and it is why treating these platforms as a career rather than a supplement has always been risky. The unpaid time made the arithmetic worse everywhere. Workers spent an average of 20 minutes on unpaid activities for every hour of paid work, searching for tasks, taking unpaid qualification tests, researching clients to avoid fraud and writing reviews. Task scarcity drove that. Fifty eight percent of workers reported that the availability of tasks was insufficient, and 88 percent said they would like to do more crowdwork, wanting 11.6 more hours a week on average. That is a workforce that was underemployed before this closure, not one that is losing a full schedule now. Metaintro has covered a related squeeze in [how AI compresses wages for millions of workers](https://www.metaintro.com/blog/ai-wage-compression-58-million-workers-2026) without formally eliminating their roles. ## Who actually depended on this income? The flexibility mattered to specific groups in ways a headline about gig work tends to flatten. The ILO survey found that ten percent of respondents had health conditions that affected the type of paid work they could do, and that for many of them crowdwork was a way to keep earning at all. Care responsibilities split sharply by gender, with 13 percent of women workers saying they could only work from home for that reason compared with 5 percent of men. Pawloski's own path illustrates the pattern. She started on the platform in 2008 while on maternity leave for supplemental income, then moved to it full time after being let go from her job in 2012 to help care for her son with special needs, and began organizing with Turkopticon in 2019. She described the work as offering "meaningful income" and flexible hours comparable to ride hail driving or [Amazon Flex](https://flex.amazon.com/) delivery, with the added benefit of being remote and completable on a smartphone. "There's some people that still pretty much still do it full time," she said. "They're concerned now." Some insurance and travel companies still relied on the platform for data workers and are now scrambling to find alternatives, which means demand did not vanish along with the marketplace. ## Is AI training work still a real career path? Yes, but the shape of it has changed and the entry point has moved up. The reason Mechanical Turk lost ground is that buyers of training data wanted specialists rather than an open crowd. That shift rewards workers who can demonstrate domain knowledge, consistent quality and the ability to explain a judgment call, which is a different skill set from clearing high volumes of cheap tasks. The growth is now in oversight rather than raw labeling. Metaintro has tracked the rise of roles built around checking machine output in [the hours workers lose to botsitting](https://www.metaintro.com/blog/botsitting-ai-oversight-hours-2026) and in [how managing AI agents became its own job](https://www.metaintro.com/blog/coming-burnout-from-managing-ai-agents). Those roles pay for accountability, not for speed. The same logic is playing out in adjacent fields, as [AI takes over freight quoting](https://www.metaintro.com/blog/ai-freight-quoting-broker-dispatcher-jobs-2026) while the humans who remain move into exception handling. Workers who spent years developing an eye for what a bad label looks like hold a genuinely transferable asset, provided they can describe it in the language a hiring manager understands. ## What should you do if this was your income? Start by documenting the work, because most turkers have no employment record to show for years of effort. Write down task categories handled, approval rating, volume completed and any specialist qualifications passed. That record is the raw material for a resume that reads as quality control and data annotation experience rather than as a gap. Workers who treated the platform as a full time job have real production data behind them, and it is worth extracting before the site goes dark on September 30. Next, widen the target beyond annotation. The transferable core is careful remote work performed to a standard without supervision, which maps onto content moderation, transcription, quality assurance, claims review and research support. Metaintro's guides to [jobs that do not follow a nine to five](https://www.metaintro.com/blog/jobs-outside-9-to-5) and [what front line workers actually want from AI at work](https://www.metaintro.com/blog/front-line-workers-fine-with-ai-want-transparency) are useful starting points for framing that pitch. If an offer arrives with a fixed rate, remember that pay is not the only lever, as Metaintro sets out in [seven things you can negotiate when the salary will not move](https://www.metaintro.com/blog/things-to-negotiate-besides-salary-2026). Finally, apply the lesson about platform risk. Amazon gave roughly a month of public notice after quietly closing the door to new customers. Any single platform holding your entire income can do the same. Workers who spread across several buyers, or who convert a platform relationship into direct client work, absorb that kind of announcement far better than workers who do not. Metaintro has documented how thin that margin can be for people working under a single employer's terms in [its reporting on Amazon delivery driver pay](https://www.metaintro.com/blog/new-jersey-sues-amazon-delivery-driver-wages-2026). ## What does this mean for your career? The closure is a signal about where human work sits in the AI economy, and it is not the signal most headlines will draw. Human input is not being removed from AI systems. It is being professionalized. The open crowd model that paid a median of about 2 dollars an hour is losing to arrangements that vet workers, pay more and expect more, because model quality now depends on data quality in a way it did not in 2005. For anyone whose income touches this category, the move is to climb from task work to judgment work. That means specializing in a domain, keeping evidence of accuracy and reliability, and looking for employers who treat human review as a function rather than a cost to be minimized. The workers most exposed today are those whose only credential is a volume count on a platform that will not exist in a month. The workers best positioned are those who can prove they catch what the model gets wrong. That capability is scarce, it is getting more valuable, and unlike a marketplace listing it belongs to the worker rather than to the platform. ## Related Articles - [Even AI Research Jobs Are Not Safe After Amazon's Latest Round](https://www.metaintro.com/blog/even-ai-research-jobs-not-safe-amazon-cuts) - [Amazon Workers on Federal Aid Tripled Since 2020](https://www.metaintro.com/blog/amazon-workers-on-federal-aid-tripled-since-2020) - [AI Is Hiding a Learning Debt That Could Stall Your Career](https://www.metaintro.com/blog/ai-hiding-learning-debt-could-stall-your-career) - [Why Job Switchers Got 7% Raises in a Month the US Lost 23,000 Jobs](https://www.metaintro.com/blog/job-switchers-7-percent-raises-us-lost-23000-jobs-july-2026) - [OpenAI's 30 Billion Dollar Georgia Data Center and the Jobs It Will Create](https://www.metaintro.com/blog/openai-georgia-data-center-jobs-it-will-create) - [Only 19 Percent of Junior Staff Say They Have the AI Tools They Need](https://www.metaintro.com/blog/ai-access-gap-junior-staff-tools-2026) - [Game Studios Are Shrinking to One-Person Teams as AI Takes the Work](https://www.metaintro.com/blog/game-studios-shrinking-to-one-person-teams-ai) - [Why Workers Say Their Jobs Now Feel Like a Situationship](https://www.metaintro.com/blog/why-workers-say-jobs-feel-like-a-situationship) - [Texas Halts New Data Center Grid Connections and Thousands of Construction Jobs Wait](https://www.metaintro.com/blog/texas-data-center-grid-audit-construction-jobs-2026) ## People Also Asked ### Q: When exactly does Amazon Mechanical Turk shut down? A: The platform closes on September 30, 2026, according to a notice Amazon posted to the Mechanical Turk website and reported by CNBC. Amazon had already stopped accepting new customers about a month before confirming the closure. Workers with pending tasks or unwithdrawn balances should resolve them before that date rather than assuming access continues afterward. ### Q: How much did Mechanical Turk workers actually earn? A: One analysis of 2,676 workers completing 3.8 million tasks found a median wage of roughly 2 dollars an hour, with only 4 percent of workers clearing 7.25 dollars an hour. The International Labour Organization found median earnings of 2.16 dollars an hour across five microtask platforms once unpaid working time was counted. Individual tasks typically paid a few cents each. ### Q: Is there still work for people who trained AI models on crowdwork platforms? A: Yes. Demand for human training data continues, and specialist firms including Scale AI and Prolific recruit workers for exactly this purpose. Some insurance and travel companies that relied on Mechanical Turk are actively looking for alternative platforms. The difference is that these buyers tend to vet workers and expect demonstrable domain knowledge rather than volume alone. Losing a platform is not the same as losing a skill, and the workers affected here have years of documented accuracy behind them. [Metaintro](https://www.metaintro.com) surfaces verified remote and data roles from employers who are actively hiring, so that experience can go somewhere that treats it as a qualification. [Create a free profile](https://www.metaintro.com/signup) and put those years of task work in front of companies that need people who catch what the model gets wrong. ### Share this article [https://x.com/intent/tweet?text=Amazon%20Shuts%20a%2021%20Year%20Old%20Gig%20Work%20Platform%20on%20September%2030%20and%20What%20Its%20Workers%20Do%20Next&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026](https://x.com/intent/tweet?text=Amazon%20Shuts%20a%2021%20Year%20Old%20Gig%20Work%20Platform%20on%20September%2030%20and%20What%20Its%20Workers%20Do%20Next&url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026)[http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026](http://www.facebook.com/sharer.php?u=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026)[https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026&title=Amazon%20Shuts%20a%2021%20Year%20Old%20Gig%20Work%20Platform%20on%20September%2030%20and%20What%20Its%20Workers%20Do%20Next](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026&title=Amazon%20Shuts%20a%2021%20Year%20Old%20Gig%20Work%20Platform%20on%20September%2030%20and%20What%20Its%20Workers%20Do%20Next)[mailto:?subject=Amazon%20Shuts%20a%2021%20Year%20Old%20Gig%20Work%20Platform%20on%20September%2030%20and%20What%20Its%20Workers%20Do%20Next&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026](mailto:?subject=Amazon%20Shuts%20a%2021%20Year%20Old%20Gig%20Work%20Platform%20on%20September%2030%20and%20What%20Its%20Workers%20Do%20Next&body=https%3A%2F%2Fwww.metaintro.com%2Fblog%2Famazon-mechanical-turk-shutdown-gig-workers-2026) ![](https://cdn.metaintro.com/rs:fill:1200:800/q:30/plain/images/bridges/bridge-expand.1df895c6bd76d96f.png) For job seekers ## Ready to find a role that actually fits? Upload your résumé, start a Job Search Thread, and let Metaintro rank real openings against your experience — then guide you from search to offer. [Get Started Free](/signup)[Search matching jobs](/jobs/search) Match Compare live roles against your current evidence. Position Turn proof projects into role-specific applications. Improve Use market feedback to keep the skill plan current. [Return to navigation](#main-navigation)