The Notice Itself Is Short
Amazon's wording is the routine kind: the company regularly evaluates its programs, tools and services and makes adjustments accordingly, and "following an assessment, we've made the decision to close AWS Mechanical Turk, effective September 30, 2026." The notice states the platform will permanently close and points workers and requesters to an FAQ page on how to prepare. No further reasoning is given. On the timeline, this was not sudden. New customer signups closed on July 30, and the same batch shut AWS SageMaker Ground Truth and Amazon Augmented AI — two other products serving data labeling and human review. What Amazon is retiring is not one platform but a whole product line for wiring people into machine learning workflows.
The Name Always Said What This Was
MTurk launched in 2005 to break work that computers could not do, but humans found easy, into "Human Intelligence Tasks" paying a few cents each: labeling data, transcribing audio and video, filling out surveys. Bezos characterized it as "artificial artificial intelligence." The name came from the famous 18th-century chess-playing automaton — apparently a machine, actually a person concealed in the cabinet. That metaphor reads sharply today. For much of the past two decades, a great deal of what was sold as AI had people sitting underneath it. Models are now capable enough that far fewer of them need to sit there.
Why It Declined, and One Irony
Krista Pawloski, a data worker and organizer with the advocacy group Turkopticon, says MTurk had been in decline for years: models advanced quickly while data startups such as Scale AI, Mercor and Prolific moved in to recruit workers, and Amazon appeared to invest fewer resources in the service as competitors gained ground. Demand shifted and competition arrived at the same time. One irony is worth recording. A 2023 study by Swiss researchers found that in one experiment up to 46% of MTurk workers appeared to be using AI models to complete their tasks. A platform built on "find a human when the computer cannot" ended up with nearly half its humans using computers. For teams still using it, the action is concrete: roughly five weeks remain to move labeling and evaluation workflows to Scale AI, Mercor, Prolific or similar, and to confirm historical data is exported. The notice says permanent closure, not a pause on new work.
via: CNBC, The Next Web, Quartz