Remote Opportunities
Article by
Mindrift Team

Amazon is shutting down Mechanical Turk (MTurk) permanently on September 30, 2026. The closure was announced on August 25, 2026, after new requester registrations stopped in July. Contributors have until the deadline to verify payment details and complete final tasks, and many are already moving to AI training platforms like Mindrift.
After 21 years, one of the oldest names in online microtask work is closing for good. With MTurk shutting down on September 30, 2026, hundreds of thousands of people who earned money completing small digital tasks now need a new place to apply their time and attention.
The good news is the market has not disappeared – it’s just changed shape. Demand for human judgment has moved from cent-per-task marketplaces to AI training, where contributors help evaluate and improve AI models instead of filling out endless forms.
This guide walks through what Mechanical Turk was, the full timeline from its 2005 launch to its 2026 closure, why Amazon pulled the plug, and what former Turkers can realistically do next – including how Mindrift projects work and who can join.
Ready to make your next move?
Explore all-access and specialized projects on Mindrift!
What was Amazon MTurk?
Amazon Mechanical Turk launched in 2005 as a crowdsourcing marketplace operated under Amazon Web Services. Businesses and researchers, known as requesters, posted small digital jobs called Human Intelligence Tasks (HITs) – labeling images, transcribing audio, categorizing products, or answering surveys. Workers, often called Turkers, picked up these tasks and earned a fee per completed HIT, typically a few cents each.
The platform took its name from an 18th-century chess-playing "machine" that secretly hid a human chess master inside. The reference was deliberate. Amazon founder Jeff Bezos famously described the service as "artificial artificial intelligence" – software could send a task to the platform through an API and receive a result, while actual humans did the thinking behind the scenes.
For two decades, MTurk quietly powered a surprising share of the digital economy. Academic researchers recruited study participants there, companies cleaned and categorized their data there, and early machine learning teams sourced their labeled training data there. At its peak, Amazon reported the platform served more than 500,000 workers worldwide.
The MTurk timeline from 2005 to 2026
The story of MTurk tracks the story of human data work itself – from novelty, to infrastructure, to disruption by the very AI systems it helped build. The milestones below show how the platform rose and why its closure fits a larger pattern.
Year | Milestone |
|---|---|
2005 | MTurk launches under Amazon Web Services as a marketplace for Human Intelligence Tasks (HITs) |
2007 | Jeff Bezos popularizes the phrase "artificial artificial intelligence" – the platform reports more than 100,000 workers in over 100 countries |
2008 | Worker advocacy takes shape. Tools like Turkopticon let Turkers rate requesters and flag unfair pay |
2010s | MTurk becomes a standard source of study participants for academic research and labeled data for early machine learning, serving more than 500,000 workers at its peak |
2023 | Researchers at EPFL estimate 33 – 46% of MTurk workers used large language models to complete a text summarization task, raising doubts about data quality |
July 2026 | Amazon quietly stops accepting new MTurk requesters – many workers read it as a warning sign |
August 25, 2026 | Amazon announces the closure following an internal assessment, as first reported by CNBC |
September 30, 2026 | MTurk permanently closes – the MTurk worker option also disappears from SageMaker Ground Truth and Amazon Augmented AI |
The last two rows matter most for anyone still active on the platform. The September 30 deadline is final, and it applies to the entire MTurk-connected ecosystem, not just the marketplace itself.
Why is MTurk shutting down?
Amazon's official explanation was brief – the company said it regularly evaluates its programs and decided to close MTurk after an internal assessment. Behind the corporate phrasing, three longer trends explain the decision.
Modern AI absorbed the simple tasks
The microtasks MTurk was built on – basic image labeling, short transcriptions, simple categorization – are exactly the tasks today's AI models handle cheaply and instantly. The work computers "couldn't do yet" in 2005 is now largely automated.
Data quality became harder to guarantee
The 2023 EPFL study cited above found a large share of workers were quietly using AI tools to complete text tasks. For requesters paying for human judgment, that undermined the platform's core promise – and MTurk's open, anonymous crowd model offered few tools to detect it.
The market moved upmarket
A new generation of AI data companies built managed platforms with vetting, quality controls, and higher pay for skilled contributors. As AI labs shifted budgets from bulk labeling to expert feedback, investment followed – and MTurk, largely unchanged for years, was left serving a shrinking niche.
None of this means human input stopped mattering. It means the market now rewards a different kind of human input. Think: judgment, expertise, and careful evaluation, rather than high-volume clicking.
What the closure means for workers and requesters
For workers, the immediate priorities are practical. Before September 30, 2026, confirm your payment method and transfer schedule in the MTurk worker portal, complete or return any pending HITs, and save any records you may need, such as earnings history for tax purposes. Amazon has published a closure FAQ, but once the platform goes dark, unresolved balances become much harder to chase.
The financial impact is real. While MTurk pay was famously low, advocacy groups note some workers relied on it as meaningful, flexible income – including people for whom remote microtask work was a primary earning option. Losing that income with roughly a month's notice is a genuine hardship, and it explains why migration threads have taken over communities like r/mturk.
Requesters face a continuity problem of their own. Companies and research teams that built pipelines on MTurk – or on the MTurk worker pool inside SageMaker Ground Truth and Amazon Augmented AI – need a replacement workflow before the deadline. Amazon has said requester balances will be refunded, but the workflows themselves must move elsewhere.
Where can MTurk workers go next?
The instinct after a platform closure is to find an identical replacement. A better strategy is to follow where the demand actually went.
The companies that once paid for bulk microtasks now pay for AI training – human feedback that teaches models to be more accurate, useful, and safe. The skills that made someone a reliable Turker transfer well: attention to detail, following task instructions precisely, and consistent quality across repetitive work.
Platforms like Mindrift connect contributors with AI training projects run for leading AI companies. Mindrift itself is operated by Toloka AI, a global AI data company working in this field since 2014 and headquartered in Amsterdam.
Unlike aggregator sites that simply list links to other platforms, Mindrift runs our own projects – contributors complete tasks directly on the platform and receive payment for completed tasks there too.
Some projects are open to all skill levels and feel familiar to anyone with microtask experience – structured, instruction-driven, and doable in short sessions. Others are specialized projects where professional expertise in coding, writing, law, or STEM earns significantly higher rates. Contributors can start with non-specialized tasks and qualify for specialized projects over time.
Ready to make your next move?
Explore all-access and specialized projects on Mindrift!
What can you do on Mindrift?
Mindrift projects vary as AI companies launch new training initiatives, but most tasks fall into a few recognizable categories. Former MTurk workers will find the format familiar even where the content is new.
Annotation tasks: Label and review images, video, and text so AI models learn from accurate examples. Dedicated annotation projects are the most natural starting point for experienced microtask workers.
Data collection: Record short audio or video clips, including first-person footage, following clear specifications – no professional equipment or prior experience required. Explore video data collection projects here.
Prompt writing and evaluation: Write prompts that challenge AI models, then rate and compare the responses against quality guidelines.
Specialized expert projects: Contributors with professional backgrounds can qualify for domain projects in coding, STEM, writing, legal, and consulting, where rates can reach up to $90 per hour depending on the domain. Explore specialized projects now.
Every task shows the rate before you accept, and payments are processed for completed tasks rather than hours logged – you can read exactly how Mindrift pays before you start.
Rates differ by task complexity, and it helps to understand why non-specialized tasks pay less than specialized projects when planning what to aim for.
One honest caveat: Mindrift is not a HIT marketplace with an endless task queue.
Tasks are organized into projects, and availability depends on which projects are active for your profile, skills, and language. Contributors set their own schedules, participate as much or as little as they choose, and remain free to use other platforms at the same time.
How to move from MTurk to Mindrift
There are two ways to join the Mindrift platform:
Register for all-access tasks
All-access tasks are open to every contributor on the platform and typically don’t require specialized domain knowledge or extensive experience. To register:
Visit the registration portal
Choose a registration method
Fill out some basic information
Start browsing all-access tasks
Learn more about all-access tasks, including what they are and how much they pay.
Specialized projects for experts
Specialized, or domain, projects require extensive knowledge and experience within a specific domain like mathematics, design, business, law, etc. This method involves more effort but joining gives you access to higher-paid projects.
Browse open opportunities
Submit a CV to an opportunity that matches your knowledge, skills, and experience
If your profile matches a project, complete a short screening assessment for the projects you want to join
Once you pass, get access to the platform, review project guidelines, and try your first tasks
Complete tasks at your own pace and receive payment for approved submissions
There is no need to wait for MTurk's final day. Setting up your Mindrift profile now means your screening is done before your MTurk income actually stops, so the transition happens smoothly.
Frequently asked questions
When is MTurk shutting down?
Amazon Mechanical Turk closes permanently on September 30, 2026. Amazon announced the decision on August 25, 2026, after stopping new requester registrations in July. The same date removes the MTurk worker option from SageMaker Ground Truth and Amazon Augmented AI.
Why is Amazon closing Mechanical Turk?
Amazon cited an internal assessment of its programs and services. The broader context includes AI models automating the simple tasks MTurk specialized in, research showing many workers used AI tools to complete tasks, and competition from managed AI training platforms offering better quality controls and higher pay.
Is Mindrift a good alternative to MTurk?
Mindrift suits MTurk workers who want a structured, remote, task-based gig – with a shift from bulk microtasks toward AI training tasks like annotation, data collection, and response evaluation. It is not an identical HIT marketplace, and task availability varies by project, but rates are shown upfront with a possibility of qualifying for higher-paying specialized projects.
Do I need experience to join Mindrift?
No AI or machine learning experience is required. Some tasks are open to all skill levels, including annotation and data collection, while specialized projects require professional expertise in fields like coding, writing, or law. A screening assessment matches you to suitable projects.
How does Mindrift pay compared to MTurk?
Mindrift pays per completed and approved task, and every task displays its rate before you accept it. Non-specialized tasks pay modest rates, while specialized expert projects can pay up to $90 per hour depending on the domain. As with any project-based platform, earnings depend on task availability and the projects you qualify for.
Is Mindrift legitimate?
Yes. Mindrift is operated by Toloka AI, an established AI data company working with leading AI labs since 2014 and headquartered in Amsterdam. Contributors work as independent freelancers, choose their own schedules, and receive payment for completed tasks through the platform.
Make your move to Mindrift
MTurk's closure ends an era, but the demand for human judgment behind it is stronger than ever – it has simply moved to AI training. Former Turkers bring exactly the habits these tasks reward: precision, consistency, and the ability to follow detailed instructions.
Mindrift offers a direct way to put those habits back to work – flexible, remote, project-based tasks with rates shown upfront, a path from open-to-everyone tasks to specialized expert projects, and a global community of over 20,000 contributors.
Apply to Mindrift now, complete your screening before the MTurk deadline, and start your first AI training tasks without missing a payment cycle.
Want more great reads? Check these out:
FAQ: A new way to join Mindrift — open to everyone
Debunked: Do you need a computer science degree to train AI?
Article by

Mindrift Team


