Amazon is closing Mechanical Turk on September 30, 2026, along with the companion labeling tool SageMaker Ground Truth, retiring the marketplace that Jeff Bezos once branded "artificial artificial intelligence" after twenty one years online. It's a fitting bookend: MTurk built the business of paying anonymous humans to do AI's grunt work, and it's shutting down just as the industry it seeded has moved on to better funded successors.

  • Amazon will shut down both Amazon Mechanical Turk and SageMaker Ground Truth on September 30, 2026, ending MTurk's 21 year run since its 2005 launch.
  • MTurk once had more than 500,000 workers completing small paid tasks called HITs: labeling, transcription, surveys, and other work that AI systems needed humans to do.
  • A 2023 study estimated that 33 to 46 percent of MTurk workers were already using large language models to complete writing tasks, eroding the platform's value as a source of unfiltered human data.
  • Scale AI, Mercor, and Prolific have taken over the market MTurk pioneered, supplying the human feedback data that trains today's frontier AI models.
Timeline of AI's human data pipeline, 2005 to 2026A horizontal timeline showing Mechanical Turk's 2005 launch, a 2023 study on LLM contamination of its data, the 2026 rise of Scale AI, Mercor, and Prolific, and MTurk's September 30, 2026 shutdown. WHO TRAINS AI'S DATA, 2005-2026 2005 MTurk launches "artificial artificial intelligence" 2023 Study: LLM use taints MTurk data 2026 Scale AI, Mercor, Prolific dominate RLHF labeling SEP 30, 2026 MTurk + Ground Truth shut down Platform or market shift Research finding genztech.blog
Fig 1 Mechanical Turk opened AI's human data pipeline in 2005. A 2023 study found its own workforce leaning on LLMs, and by 2026 Scale AI, Mercor, and Prolific run the market it started.

What is Amazon actually shutting down?

Two products go away on the same day. Mechanical Turk is the marketplace itself: a site where "requesters" posted small paid jobs called HITs, Human Intelligence Tasks, and anonymous workers picked them up for anywhere from a few cents to a few dollars apiece. SageMaker Ground Truth is the newer, AWS native cousin, a managed data labeling tool that let companies route annotation work, bounding boxes on images, transcript cleanup, text classification, through automated models or human reviewers, some of whom came from the MTurk pool. Both close on September 30, 2026. Amazon has not announced a replacement, and nothing in its public statements suggests one is coming.

RelatedCloudflare says 52% of crawler traffic is AI training

Why is Amazon shutting down Mechanical Turk now?

Amazon's own explanation is thin. The company says the decision followed an internal assessment of its programs and services, without laying out which metrics tipped the scale or what the review actually found. AWS has not published usage figures, worker counts, or a cost breakdown, so the "why now" has to be pieced together from context. MTurk turned 21 this year, ancient by internet platform standards, and it spent much of the last decade coasting on inertia rather than growth. The market it once had almost entirely to itself, matching AI developers with human labor, now has serious, well capitalized competitors that MTurk was never built to compete with.

What replaces Mechanical Turk?

Scale AI, Mercor, and Prolific are the names doing the work MTurk used to do, and they are doing it for a different kind of customer. Where MTurk was a self serve marketplace anyone could post a task to, these newer platforms sell curated pipelines: vetted annotators, domain specialists, and structured workflows built specifically for reinforcement learning from human feedback, the technique labs use to align chatbots and coding models. That is a more specialized, higher margin business than paying gig workers a few cents per HIT, and it is where the venture money has gone. MTurk pioneered the idea that AI needs humans in the loop. It just never turned into the company that profits from that idea at today's scale.

  1. 2005Amazon launches Mechanical Turk, letting requesters post small paid HITs for anonymous online workers.
  2. 2023A study estimates 33 to 46 percent of MTurk workers are using LLMs to complete writing tasks, undercutting the platform's value as unfiltered human data.
  3. 2026Scale AI, Mercor, and Prolific become the dominant suppliers of human feedback data for training frontier AI models.
  4. Sep 30, 2026Amazon shuts down both Mechanical Turk and SageMaker Ground Truth, ending MTurk's 21 year run.

Why did LLM contamination undermine MTurk's usefulness?

There is a more specific problem baked into MTurk's decline, and it is almost poetic. A 2023 study estimated that somewhere between 33 and 46 percent of MTurk workers were already leaning on large language models to complete writing tasks on the platform. If a company posted a HIT asking workers to write product descriptions or answer open ended questions in their own words, a meaningful share of those "human" answers were partly AI generated. That is fine for some jobs. It is disqualifying for the one job MTurk data was most valuable for: training AI models on genuinely human judgment and language. Once part of your human data pipeline is quietly relaying answers from the same kind of model you are trying to train, the data stops being an independent signal.

What does this shutdown mean for AI training data?

The industry's answer to that contamination problem has been more structure, not less: identity verification, task specific qualification tests, smaller pools of vetted specialists instead of an open crowd. Scale AI, Mercor, and Prolific all sell some version of "we know who actually did this work," a harder claim to make on an anonymous marketplace built in 2005. That shift also concentrates a market that used to be genuinely open. Fewer, larger vendors now control more of the human feedback supply chain that frontier labs depend on, which raises its own questions about pricing power and worker conditions even as it solves the trust problem MTurk never fixed.

RelatedFrance's top court kills the under-15 social media ban

What to watch: whether the RLHF labeling boom at Scale AI, Mercor, and Prolific keeps consolidating into a handful of well capitalized vendors, whether any of them run into their own version of MTurk's LLM contamination problem as their workers reach for AI tools too, and whether Amazon quietly re-enters this market under new branding once it decides it needs an in house data pipeline again.

Our take

There is something almost tidy about MTurk closing the same year the AI industry is having its loudest arguments about data quality and provenance. Amazon built a genuinely important piece of internet infrastructure with Mechanical Turk, then let it stagnate while the market it created matured past it. Bezos's "artificial artificial intelligence" line was clever branding for a marketplace that, for a long time, actually was the plumbing behind a lot of real AI research. But plumbing needs maintenance, and MTurk's interface, worker protections, and quality controls barely changed in two decades while its customers' needs got far more demanding. Scale AI, Mercor, and Prolific did not out innovate Mechanical Turk so much as show up with better trust guarantees at the exact moment MTurk's own workers started undermining its core value proposition. That is not a coincidence worth mourning. It is just what happens when a category leader stops evolving.

Primary sources

Original analysis by GenZTech Team. Sources: CNBC.