The end date
Amazon has confirmed that Mechanical Turk, known to most of its users simply as MTurk, will close permanently on September 30, 2026. The company had already stopped accepting new customers a month earlier, on July 30, quietly winding the platform down well before the public announcement. Amazon's stated reason is characteristically brief: "We regularly evaluate our programs, tools and services and make adjustments based on those assessments."
That understated line closes out one of the stranger and more consequential corners of the internet economy, a marketplace that, for two decades, sat at the exact point where human labor and machine learning met.
Where the name came from
MTurk launched in 2005, named after an 18th-century chess-playing machine that toured Europe astonishing audiences with its ability to beat human opponents. The trick, revealed only later, was that a person was hidden inside the cabinet operating it the whole time. Amazon founder Jeff Bezos borrowed the reference deliberately: he described the platform, in a widely cited 2007 remark, as "artificial artificial intelligence," a system where tasks that looked automated were, underneath, being completed by real people.
Those tasks were called HITs, short for human intelligence tasks: small, disconnected pieces of work such as transcribing audio, labeling images, moderating content or filling out short surveys, often paying just a few cents apiece. Businesses and, especially, academic researchers used MTurk as a cheap, fast way to get large volumes of human judgment on demand. For much of the 2010s, it was also the default way machine learning researchers built labeled training data, making it, without much public recognition, one of the quiet foundations of the modern AI boom.
Why the platform outlived its usefulness
The same industry MTurk helped build eventually made it redundant. Two forces did the damage. First, the kind of routine labeling and transcription work that once required a crowd of humans is now something AI models can often do directly, shrinking the pool of tasks businesses needed a human marketplace for at all. Second, where human-generated data is still needed, specialized companies built specifically for AI training data, including Scale AI, Mercor and Prolific, have taken that business by offering more curated, higher-quality workforces than MTurk's open, anonymous crowd ever could.
There is also a more pointed irony in MTurk's decline: a 2023 academic study estimated that somewhere between 33 and 46 percent of MTurk workers were themselves using large language models to complete tasks that were supposed to be verifiably human-generated. A marketplace built on the premise of authentic human judgment had, by its final years, become partly automated from the inside, undermining the one property that made it valuable to AI researchers in the first place.
Krista Pawloski, an organizer with Turkopticon, a longstanding watchdog and advocacy group for MTurk workers, summed up the platform's trajectory bluntly: it had been "in decline" for some time before Amazon made the closure official.
What it means in practice
For businesses and researchers still relying on MTurk for surveys, data collection or content moderation, the closure means finding a replacement workflow before September 30. The direct successors are the specialized data-labeling platforms named above, though none replicate MTurk's breadth of use cases exactly; teams that used MTurk for social science research or general crowdsourcing, rather than AI training data specifically, may need to look at Prolific or academic-focused alternatives instead.
For the workers who relied on MTurk, many of them long-term participants who built routines and reputations around the platform over years, the shutdown removes a source of flexible, if low-paid, income with little public notice and no clearly signposted alternative of the same scale. That is the less visible side of this story: MTurk was never just data-labeling infrastructure, it was also a labor market for tens of thousands of people, and its closure is a labor story as much as it is a technology one.
More broadly, MTurk's exit marks a symbolic point in the AI industry's development. A platform explicitly built around making human labor look automated is shutting down because genuine automation, and better-organized human labor markets built specifically to serve it, finally overtook it. The "artificial artificial intelligence" of 2005 is being replaced by the real thing.
