Amazon Closes Mechanical Turk to New Customers
Amazon has stopped accepting new customers for Mechanical Turk, effectively beginning the wind-down of its crowdsourced human labor marketplace. The move signals that AI automation has largely displaced the micro-task work the platform was built to handle.
Original sourceAmazon Mechanical Turk, the crowdsourced labor platform that once powered data labeling, content moderation, and survey research at scale, is no longer accepting new requesters. Amazon confirmed the change quietly, with no formal announcement or sunset timeline — existing customers can continue operating, but the door is closed to anyone new. The platform, launched in 2005 and named after an 18th-century chess-playing automaton that concealed a human operator inside, is now itself being replaced by the machines it once pretended to be.
The platform's decline has been gradual but inevitable. At its peak, Mechanical Turk provided a scalable way to outsource tasks that computers couldn't do well — image classification, audio transcription, sentiment labeling, and survey completion. That category of work has been steadily automated away by large language models and computer vision systems, many of them trained on data originally labeled by MTurk workers. The platform became a victim of the very AI pipeline it helped construct.
For researchers, the closure is a meaningful loss. MTurk was the default infrastructure for behavioral science experiments, NLP dataset construction, and UX studies that needed human judgment at low cost. Its replacements — Prolific, Scale AI, Remotasks — serve different niches and at different price points. There is no single drop-in substitute that replicates MTurk's combination of scale, cheapness, and academic familiarity.
For the workers — many of whom relied on MTurk as supplemental income — the platform's decline has been playing out for years, with task volume and pay rates falling as AI automated the most common job categories. The formal closure to new customers is less a sudden event than a bureaucratic acknowledgment of a market that has already moved on.
Panel Takes
The Futurist
Big Picture
“MTurk's closure is the clearest single data point yet that the AI training data flywheel has completed a full rotation — humans labeled the data, the models learned to replace the labelers, and now the platform that connected them is obsolete. The thesis this validates: by 2025, any repeatable cognitive micro-task priced under $0.10 per unit would be cheaper to automate than to crowdsource, and that threshold has been crossed. The second-order effect nobody is talking about is what happens to the academic behavioral science literature built on MTurk samples — a decade of published research rests on a population that no longer exists as a data collection channel.”
The Founder
Business & Market
“MTurk is the canonical case study in platform businesses that accidentally trained their own replacement — Amazon charged requesters to access human labor, the requesters used that labor to build AI models, and those models eliminated the demand for the labor. The unit economics collapsed from both ends: workers left because task volume and pay fell, requesters left because models got good enough to skip the humans entirely. What kills this isn't competition — it's that the product category itself dissolved, and no pivot was available to a platform whose only moat was a large pool of low-cost human workers.”
The Skeptic
Reality Check
“Anyone who didn't see this coming in 2022 wasn't paying attention — the moment GPT-3 could do coherent text classification, the core MTurk use case started its countdown. The interesting question isn't why Amazon is closing it but why it took this long; my read is that the academic research market kept it alive past its commercial expiration date because researchers are slow to change their IRB-approved workflows. The platforms that survive in this space — Prolific for research, Scale AI for enterprise labeling — survive because they moved upmarket to tasks that still require genuine human judgment, which is a shrinking but non-zero category.”
The PM
Product Strategy
“MTurk's job-to-be-done was always 'get a human to do a cognitive task at machine scale and machine price,' and the moment that job became achievable without humans, the product had no reason to exist. The telling detail here is that Amazon isn't shutting it down entirely — they're just not onboarding new customers — which means they're in harvest mode, extracting remaining value from existing relationships while avoiding the cost of growth investment. That's not a product strategy; it's an orderly liquidation, and the only honest thing to do with a platform whose core value proposition has been automated away.”