By ByteBot
Amazon has confirmed September 30, 2026 as the final shutdown date for Mechanical Turk, the 21-year-old crowdsourced labor marketplace that Jeff Bezos once called “artificial artificial intelligence.” New sign-ups closed July 30, alongside AWS SageMaker Ground Truth and Amazon Augmented AI. If annotation pipelines or research workflows still run through MTurk, five weeks remain to migrate.
Three AWS Services Going Dark Together
The September 30 date is not just for MTurk. Amazon closed three related services to new customers simultaneously on July 30:
- Mechanical Turk — full shutdown on September 30; all APIs go offline
- SageMaker Ground Truth — closed to new customers; existing accounts continue post-July 30
- Amazon Augmented AI (A2I) — also closed to new customers
For active MTurk requesters, the distinction matters. SageMaker Ground Truth existing customers keep running. MTurk is a complete close — no APIs, no marketplace, no extensions.
Why It Died: AI Ate Its Own Supply Chain
Two forces converged to make MTurk unviable. The first was straightforward market competition — Scale AI, Prolific, and Surge AI had been drawing workers and enterprise customers away for years, and Amazon put fewer resources into the platform as rivals arrived.
The second cause cuts deeper. A 2023 study found that 33 to 46 percent of MTurk workers were using large language models to complete text annotation tasks. A separate analysis found the figure reached 40 percent in text summarization work. Amazon was selling “human intelligence” that had already become machine output. Researchers collecting training data were often receiving LLM-labeled content that passed as human judgment.
The platform’s entire value proposition rested on authentic human signal. Once that eroded, the case for MTurk collapsed. The same LLMs that needed human training data had rendered the humans providing it redundant — sometimes at their own initiative.
Who Needs to Act Before September 30
MTurk served a wide range of workflows:
- ML and AI teams — image labeling, text classification, sentiment analysis, RLHF data collection
- Academic researchers — behavioral studies, surveys, A/B experiments, psychology research at scale
- Startups — content moderation, data validation, human review before automation was ready
If any of those describe active pipelines, the clock is running. Amazon has not indicated any extension to the September 30 date.
Where to Go: Migration Options by Use Case
There is no single drop-in replacement, but the alternatives are mature:
Enterprise ML annotation: Scale AI is the most capable platform available, though Meta’s 49 percent stake is a disqualifying factor for some labs. Surge AI is the alternative — elite labelers, quality over volume, lab-independent — and has become the default for RLHF work where neutrality matters.
Academic and behavioral research: Prolific is the clear replacement. Over 300,000 registered participants, a minimum $8 per hour pay rate regardless of location, and a reputation for quality free-text responses. It costs more per response than MTurk did, but the data is cleaner.
AWS-native workflows: SageMaker Ground Truth Plus keeps annotation on the AWS stack with a managed workforce option. If the rest of the ML pipeline is already in AWS, this is the lowest-friction migration path.
Sensitive data and custom needs: An in-house annotation team trades cost for control. Expensive to stand up but eliminates vendor dependency and data privacy concerns entirely.
Human-in-the-Loop Is Not Dead — It Is Shrinking
The pattern replacing MTurk-style annotation is not “no humans” — it is “fewer humans, better placed.” The economically viable model now: an LLM handles the first pass, and a smaller pool of domain experts reviews edge cases. The days of routing millions of tasks to anonymous crowdworkers are effectively over.
MTurk exits having accelerated its own irrelevance. Workers who used LLMs to complete tasks faster were rational actors responding to incentives, but they hollowed out the platform’s value from the inside. Amazon’s decision to stop investing simply acknowledged what the market had already decided.
Five weeks. Check your annotation pipelines, review your requester account status, and pick a migration target before September 30.
