The short answer
Amazon Mechanical Turk closed for good on September 30, 2026. SageMaker Ground Truth Plus, AWS’s managed labeling service, reached end of support in June. SageMaker Ground Truth and Augmented AI (A2I), its human review service, are still running for existing customers, but they stopped accepting new ones in July and AWS plans no new features for Ground Truth.
That is a lot of data labeling leaving AWS at once. If your data went through MTurk, Ground Truth, or A2I, you will need a new home for it sooner or later: somewhere your existing data and labels can move to, and where new labeling can continue.
What it means for your data
The right next step depends on how you used these services.
- You posted tasks directly on MTurk. The marketplace is gone. Requesters have until October 30, 2026 to approve or reject submitted work, after which it is approved automatically. MTurk says transaction history stays available until January 28, 2027. Save whatever results and records you still can now.
- You ran Ground Truth jobs with the MTurk workforce. Ground Truth still runs for existing customers, but the public crowd option closed with MTurk. AWS recommends moving to a private or vendor workforce.
- You ran Ground Truth with your own team or a vendor. Nothing stops working today. But a tool with no new features won’t keep up as your data and requirements change.
- You used Ground Truth Plus. It is already gone. Whoever ran your labeling there needs replacing, along with the tool.
- You used A2I to review model output. Existing workflows continue, but there is no path to grow them on AWS’s side.
In every case, you are moving three things: the data, the labels you already paid for, and the record of how they were made.
What MTurk was good at, and what it wasn’t
MTurk was a fast, self serve way to buy small amounts of human judgment, and AWS advertised its workforce as “over 500,000 independent contractors worldwide.” For simple, public, low stakes tasks, that was hard to beat.
It also had firm limits. According to AWS’s own documentation, the MTurk workforce couldn’t be used on data containing personal information, and it wasn’t supported for video frame or 3D point cloud labeling jobs. Those are exactly the data types and constraints that serious AI data processing runs into. Much of the hardest labeling work never ran on the crowd in the first place.
So the question isn’t only “where do I find another crowd?” It’s what your data actually needs: who should label it, how it gets checked, and where it lives.
What to look for in a replacement
Before you move, check any replacement against this list:
- Your existing labels. Can it bring in your MTurk results and Ground Truth output manifests, so you don’t start from zero?
- Where your data lives. Can it stay in your own cloud, or does it have to be copied into someone else’s?
- Your choice of tool. Can you keep using a labeling tool your team already knows, or switch to another third party tool while maintaining an environment of your own where your data and labels sync and live?
- Who labels it. Can you use your own team, a managed team, or both?
- Data types. Does it handle video, audio, text, 3D, and sensor data, not only images?
- Quality control. Is there consensus review, automated checks, and agreement tracked per person, the job Ground Truth’s annotation consolidation used to do?
- Human review of model output. If you used A2I, can reviewers accept, correct, or reject predictions and record why?
- Output. Do labels come back in the schema your training code already reads, versioned, with a record of who touched them?
- Size of the work. Can you start with a few thousand files, or is there a large minimum?
Use MLtwist to migrate your Ground Truth
MLtwist has migration technology and services for teams moving off MTurk and SageMaker Ground Truth. Use MLtwist to download your MTurk and Ground Truth data, including the labels and output manifests you already have, and bring it into your own environment, where it can be managed by yourself or your own instance of MLtwist. From there, you choose where your labeling goes next:
- Push it to your next labeling tool. If you have already picked a new tool, MLtwist prepares the data and sends it there. We already set up and run work in third party tools, then pull, check, and post process the results.
- Label in MLtwist. Use the MLtwist labeling platform with our team, yours, or both. Prelabeling Twists use foundation models to draft the first boxes, masks, tags, and transcripts, reviewers approve or reject each file, and every labeled file carries a Data ID Card recording where it came from and who labeled and reviewed it.
- Use a new third party labeling tool while keeping the data in your own cloud. Run the MLtwist platform in your own cloud environment, with your own database. It can still push work to a third party labeling tool, but your data and every label, as it is generated, stay in your environment.
If you used A2I to check model output, model evaluation at MLtwist covers the same kind of work: reviewers accept, correct, or reject each prediction with the reason attached, and results are reported by class rather than as one overall score.
A checklist for moving off AWS labeling
- Export now. Save results, output manifests, worker instructions, and task templates while you still have access.
- Write down your label definitions. The rules your workers followed are as valuable as the labels themselves.
- Decide where the data should live. Your own cloud, a vendor’s, or a mix. It is easier to choose now than to move twice.
- Pick a small batch to test. Run a few hundred files through the new workflow and compare the results with your old labels before moving everything.
- Decide who labels. Your own experts, a managed team, or both. Higher stakes data deserves more expert review.
- Keep the record. Make sure the new setup tracks who labeled what and which version a model was trained on.
MTurk’s customers, and Ground Truth’s, will eventually need a new home for their data. It’s worth choosing one that fits where your data is going, not only where it has been.
Learn more about data labeling at MLtwist, or get in touch about migrating your MTurk or Ground Truth data.
AWS service dates and details are as published by AWS on the pages linked above.