Amazon Mechanical Turk Is Shutting Down: Your Outsourcing Alternative - BUNCH

A 21 year old platform closes September 30. The work it used to handle still needs somewhere to go.

‍

Amazon confirmed on August 25 that Mechanical Turk will permanently close on September 30, 2026. That closes out 21 years for a platform Jeff Bezos once called "artificial artificial intelligence," built to hand off the tasks a computer still could not do reliably on its own.

‍

That name became a fair critique over time. A lot of the work posted to the platform ended up going right back to AI anyway, with workers running their assigned tasks through AI tools instead of doing the labeling themselves, which defeated the entire premise. The whole point was catching what a model still couldn't handle on its own. BUNCH runs the opposite way. Real people label, moderate and validate the data that trains AI models, not AI quietly standing in for the person meant to check its work.

‍

If your team ran labeling, moderation or transcription work through MTurk, that work needs a new home now, whether the shutdown is still days away or has already happened by the time you're reading this.

‍

What Actually Happened

Mechanical Turk launched in 2005 as a marketplace for Human Intelligence Tasks or HITs: small jobs like labeling data, moderating content, transcribing audio and video or answering surveys, posted by requesters and picked up by crowdworkers known as Turkers. At its peak the platform served more than 500,000 workers worldwide.

‍

Amazon stopped accepting new MTurk customers on July 30 and confirmed the full shutdown a month later, saying only that it regularly evaluates its programs and made the call following an internal assessment. The same closure date applies to SageMaker Ground Truth and Amazon Augmented AI too. That timing suggests Amazon isn't just retiring one product. It looks like a broader step back from human data infrastructure altogether.

‍

Coverage from CNBC points to years of decline as newer data labeling platforms drew workers and business away while AI systems took over more of the routine tasks HITs used to handle. Whatever the exact mix of reasons behind it, the practical result is the same for anyone still running work through the platform. It stops on September 30 whether a replacement is ready or not.

‍

What MTurk Actually Handled

Strip away the branding and Mechanical Turk was solving one problem: getting a task in front of a human when a computer could not reliably handle it alone. In practice that meant:

‍

• Data labeling and tagging across text, image and audio datasets
• Data validation and verification
• Content moderation
• Audio and video transcription
• Survey responses and data collection
• Basic image classification for computer vision work

‍

That work doesn't disappear just because the platform that used to host it is shutting down. It still needs to get done. It just needs somewhere new to run.

‍

Why a Managed Team Beats an Anonymous Crowd

MTurk always came with a tradeoff. Requesters got access to a huge, flexible pool of workers, but they also inherited the entire QA burden themselves. Guidelines had to be simple enough for a stranger to follow with no training, since there was no ongoing relationship with the people doing the work.

‍

BUNCH runs the opposite model. Instead of a rotating, anonymous crowd, clients get a dedicated team that actually learns their product, their edge cases and their guidelines over time. Every batch of work gets checked by a QA analyst before it goes back to the client rather than left to a requester to sample and hope for the best. Teams scale up for a launch or a volume spike and scale back down once things settle, without anyone having to rebuild a workforce from scratch each time. That consistency shows up downstream too. On a fashion segmentation project for an AI fashion company, annotators had to capture nuanced attributes precisely enough to hold up in production and the client credits that precision with improving its model's performance. Agents and reviewers are held to a minimum 90 percent SLA score, with response times capped at one minute, a bar the old crowdsourced model never set at all.

‍

If data labeling is the piece of your MTurk workload you need solved first, BUNCH's data labeling team already runs this work every day for AI and product companies. Get in touch and see what your own data looks like once it runs through a dedicated team instead of a crowd.

‍

What BUNCH Can Take Over Starting Now

If your team relied on Mechanical Turk for any of the following, BUNCH already runs teams handling this work every day for AI and product companies:

‍

• Image, text, audio and video data labeling and annotation
• Data validation, verification and cleansing
• Content moderation across text, image, video and live streams
• Audio and video transcription
• Survey and data collection projects
• Bounding boxes, tagging and other computer vision annotation work

‍

BUNCH teams already scale from a handful of specialists on a small project up to hundreds handling millions of items a month for larger clients, with an average client relationship above four years. Work runs through a dedicated team rather than a rotating pool of strangers, with a QA pass built into every batch, coverage running 24/7 rather than depending on whoever happens to be online, and an account manager who actually knows the account. Engagements can also run as a fixed scope project instead of an ongoing team, say 20,000 images delivered across three months, done once the batch is out the door. Data stays with that team under NDA instead of moving through an open, public worker marketplace. BUNCH also operates as a GDPR compliant data processor. BUNCH also ranks 3rd out of 26 companies in RankMyAI's independent Philippines rankings, current as of August 2026.

‍

Beyond Data Ops

MTurk's replacement doesn't have to stop at data work. BUNCH also runs:

‍

• Customer support across chat, phone and email for SaaS, ecommerce, fintech and crypto products, integrated with tools like Zendesk, Help Scout, Intercom, Freshdesk and Salesforce Desk
• Know Your Customer (KYC) verification for fintech, crypto and prediction market platforms
• Community and social channel moderation across Discord, Telegram and crypto communities
• Content moderation across text, image, video, live streams and comments
• AI safety work including red teaming, RLHF and model or LLM fine tuning
• AI data validation and verification, with rubric-based QA and double-pass review for sensitive outputs

‍

If your evaluation isn't limited to what MTurk used to cover, this is worth a conversation too.

‍

Questions Worth Asking Before You Pick a New Vendor

Whoever picks up the work MTurk used to handle, these are worth asking before signing anything:

‍

• How fast can a new team actually get calibrated on real data before full volume hits?
• Who reviews disagreements between annotators or moderators and how do they get resolved?
• What happens to the data during and after the engagement and who has access to it?
• Is pricing predictable at scale or does it change once volume jumps?
• What does a real QA report look like, not just a promise that QA exists?

‍

Making the Move Off MTurk

A platform shutdown is not the moment to freeze. Most teams moving off MTurk are trying to solve the same two problems at once: get a new team calibrated fast and make sure quality does not drop the moment the switch happens. That's true whether you're planning ahead of September 30 or you're only finding out about the shutdown after the fact. MTurk closing doesn't make the problem go away, it just moves the deadline earlier.

‍

BUNCH runs new engagements through a short calibration first. Send over a sample of the work that was running through MTurk and BUNCH will train a small team against it, review the results together and only then scale to full volume. That is a real quality check before committing to anything, which is more than the old model ever offered.

Whether MTurk is still running or already gone by the time you're reading this, reach out to BUNCH for a free calibration and get the work moved to a team that isn't disappearing on you.

‍

BUNCH is not affiliated with or endorsed by Amazon.