Amazon Mechanical Turk Shuts Down After 21 Years, AI Killed It

Amazon spent 21 years selling the idea that a human could pass for artificial intelligence. On September 30, it's turning that idea off for good.

What's actually closing

Mechanical Turk launched in 2005 as a marketplace for what Amazon called Human Intelligence Tasks, or HITs — tiny jobs like labeling images, transcribing audio, or answering surveys, usually paying a few cents each. The name came from an 18th-century chess-playing automaton that looked mechanical but was secretly operated by a hidden human chess master. Jeff Bezos leaned into that joke early on, describing MTurk as "artificial artificial intelligence": to a company uploading a task, it looked like a machine solving the problem, when really it was a person on the other side of a screen.

That joke aged badly. On July 30, Amazon quietly stopped accepting new MTurk customers, along with two related services, SageMaker Ground Truth and Amazon Augmented AI. Then on August 26, it made the wind-down official: all three shut down for good on September 30, and AWS is exiting human-data infrastructure as a business entirely.

The platform that trained AI got eaten by it

For years, MTurk workers were the quiet labor behind machine learning — the people labeling training data, checking model outputs, and doing the tedious annotation work that large language models eventually learned to imitate. A 2023 study out of EPFL found that somewhere between 33 and 46 percent of Turkers doing writing tasks were themselves already using LLMs to do the work, which undercuts the entire premise of the platform: paying for human judgment that's actually human.

Newer, more specialized labeling shops — Scale AI, Mercor, Prolific among them — had already pulled much of the serious enterprise data-labeling business away from MTurk's open marketplace model. Krista Pawloski, an organizer with the worker advocacy group Turkopticon, told reporters the platform had been fading for years as Amazon put fewer resources into it and workers drifted elsewhere, leaving mostly insurance and travel companies still relying on MTurk — and those remaining customers are now scrambling to line up replacements before the shutdown.

Amazon's own explanation stayed corporate and non-specific:

"We regularly evaluate our programs, tools, and services and make adjustments based on those assessments. Following an assessment, we've made the decision to close AWS Mechanical Turk, effective September 30, 2026."

It's the kind of line that confirms a decision was made without saying why.

What workers and requesters are left to do

There's no severance package, no transition program, and no referral to a sister service. Anyone still depending on MTurk has about five weeks between the announcement and the shutdown date to sort out what comes next:

• Requesters need to export any HIT data, results, and worker records before September 30.
• Businesses still routing labeling or survey work through MTurk have to find and migrate to a new vendor.
• Workers with pending balances need to confirm payment settlement before the platform goes dark.
• Anyone using SageMaker Ground Truth or Augmented AI for human review loops in AWS pipelines needs a separate replacement for those too.

The bigger tell

What's notable isn't that a niche gig platform is closing — it's which one, and why. Mechanical Turk wasn't just another crowdsourcing site; it was one of the original scaffolds the machine-learning era was built on, quietly feeding labeled data into models for two decades. Amazon shutting it down at the exact moment AI can generate convincing text on its own isn't a coincidence, it's the natural end point. The tool that made "human-in-the-loop" AI possible is being retired because the loop increasingly doesn't need a human in it, at least not one Amazon is willing to keep paying by the task.