Amazon is shutting down Mechanical Turk after 21 years

What this is about

Amazon will close Mechanical Turk on September 30, 2026. CNBC published excerpts from a customer notice on August 25. The move ends a 21-year-old platform that matched people around the world with tiny digital jobs: labeling images, transcribing audio, answering surveys, or checking datasets.

The timing is notable. Mechanical Turk was long an invisible human layer behind automated systems. Language models can now perform some of those tasks themselves, while specialized data companies organize more complex human evaluations for AI labs.

What Mechanical Turk actually does

Requesters publish Human Intelligence Tasks. Workers choose individual jobs and receive a small payment when their work is accepted. The system breaks large projects into many small units and distributes them to an open workforce.

Amazon had already stopped accepting new customers on July 30, 2026. According to the published notice, existing requesters have about five weeks to finish ongoing work and secure their data. Amazon did not publish a detailed explanation; the notice only refers to an internal assessment of programs and services.

Why it matters

Mechanical Turk served at once as a labor market, research infrastructure, and data factory. Researchers recruited study participants there, companies had content sorted, and early machine-learning systems received human-produced labels. Its closure therefore affects more than an aging AWS product. It removes one part of the supply chain on which modern AI was built.

There is an irony in the shift. A 2023 EPFL study covered by TechCrunch estimated that 33 to 46 percent of the workers examined were already using language models for text tasks. That estimate applied to one study and must not be generalized to every task or worker. It nevertheless illustrates how difficult it had become to separate human answers from machine-generated ones.

In plain language

Mechanical Turk was like a huge kitchen where thousands of people chopped vegetables for a machine that would eventually learn to cook. Once the machine could handle some chopping, requesters could not always tell whether a person or the machine was at the cutting board. At the same time, new kitchens appeared with tighter controls over who handles each demanding task.

A practical example

A research team wants human reviewers to assess 10,000 text answers in September 2026. At 20 cents per task, it would budget $2,000 in worker payments alone, before platform fees. Because of the announced closure, the team must now determine whether it can finish the project in time.

If it switches providers, it cannot simply combine all answers. Participant selection, pay, quality controls, and consent may differ. For a scientific study, changing platforms can therefore also change the comparability of the data.

Scope and limits

• Public reporting relies on a customer notice; Amazon has not published a detailed business rationale.
• The closure does not prove that human data work is disappearing. It is moving to specialized platforms, internal teams, and subject-matter experts.
• The EPFL estimate of 33 to 46 percent applies to the studied scenario, not the entire Mechanical Turk workforce.

It is also unclear how many active workers and requesters are directly affected. Amazon did not provide current figures in the notice. Only the end date is firm: existing workflows must be moved by September 30, 2026.

SEO & GEO keywords

Amazon Mechanical Turk, MTurk, platform labor, data annotation, AI training, Human Intelligence Tasks, crowdwork, future of work, AWS, language models, research data