Amazon Mechanical Turk shuts down for good on September 30, 2026, after 21 years, and it takes the crowd workforce behind SageMaker Ground Truth and Amazon Augmented AI with it. If your labeling pipeline runs on either, you have about two weeks to get your data out and a new pipeline running.
Where do you go now? Roboflow is a great Mechanical Turk plus SageMaker Ground Truth and Amazon Augmented AI alternative. This post covers what is actually ending, what to do before the deadline, and how to move a Ground Truth project into Roboflow without relabeling.
Mechanical Turk Worker Type Will No Longer Be Available
Ground Truth Plus, the managed labeling service, reached end of support on June 30, 2026. Ground Truth, Augmented AI, and Mechanical Turk all stopped accepting new customers on July 30. And on September 30, MTurk closes permanently.
The Mechanical Turk worker type will no longer be available when creating labeling jobs or human review workflows as of that date.
Ground Truth is closed to new customers, with no new features coming, and its public workforce option is gone.
Your Migration Checklist Before September 30
First, export your output manifests. Every Ground Truth job writes its labels to an output manifest in S3, one JSON line per image, with the S3 path, the bounding boxes, and the class map. Copy those manifests and the source images somewhere you control. Do this today, not on the 29th.
Second, close out your HITs. Submissions stop on September 30. You then have until October 30 to approve or reject submitted work and to pay bonuses. Anything you don't review within 30 days auto-approves, and Amazon refunds prepaid balances within 30 days of closure. Transaction history stays available until January 28, 2027.
Bring Your Manifests to Roboflow
Roboflow reads the SageMaker Ground Truth manifest format natively. Upload the manifest alongside your images and your existing bounding boxes land in a Roboflow project, ready to review, edit, or export to COCO JSON, YOLO, Pascal VOC, or any of 25+ other formats.
Every label you add from here on lives in one dataset with version history in Roboflow, with health checks and augmentation built in.
Use Auto Label in Roboflow
Ground Truth's design assumed the only way to label at scale was to send images to hundreds of people and average their answers. That's no longer true.
Today the first pass of labeling is done by a foundation model. Auto Label in Roboflow runs GPT-6 Astra and Gemini over your dataset and drafts the boxes and polygons for you. Your team then reviews, corrects, and approves.
Smart Polygon, powered by Meta AI's Segment Anything, turns a click into a pixel-accurate mask for the cases a model gets close but not perfect. And if you already have a model, Label Assist uses it to pre-annotate the next batch so each round gets faster than the last.
The people doing the reviewing are the people who know the product: your QA engineer who can tell a scratch from a smudge, your ops lead who knows which pallet configuration is out of spec.
From a Labeling Job to a Vision System
Labels were only ever the first step. Once your data is in Roboflow you can train RF-DETR or fine-tune from a checkpoint, deploy to the cloud or the edge, and serve the trained model over MCP.
More than 2,000,000 developers and half the Fortune 100 build on the Roboflow platform,
Bring your manifests, start a free project, and if you're moving a large or sensitive dataset, talk to us and we'll get you set up before the 30th.
Cite this Post
Use the following entry to cite this post in your research:
Erik Kokalj. (Jul 1, 2026). Mechanical Turk Closes: How to Move Labeling to Roboflow. Roboflow Blog: https://blog.roboflow.com/migrate-sagemaker-ground-truth-mechanical-turk/