Amazon is closing Mechanical Turk — trust your data less
Amazon Mechanical Turk stops taking new customers July 30. The reason — bots faking human work — is a warning about every data pipeline you trust.
For two decades, Amazon Mechanical Turk was the invisible human layer under machine learning — the marketplace where pennies-per-task workers labeled the data that trained your models and checked their outputs. On July 5, Amazon confirmed it's shutting the door: no new customers after July 30, 2026. The quiet part is the reason. The humans stopped being human.
What actually happened
AWS will stop accepting new Mechanical Turk customers on July 30, 2026, per TechCrunch. Existing customers keep access, but there are no new features planned — the service that launched in 2005 is on maintenance-only. AWS cited "careful consideration" and ongoing investment in security and availability, without a roadmap.
The uncomfortable backdrop: a 2023 study found that between 33% and 46% of Mechanical Turk workers were using large language models to complete the very tasks meant to produce human-generated data. The platform built to supply ground truth had been quietly poisoned by the models it was supposed to help train. The official notice sits on mturk.com today.
Why it matters for your business
Two lessons, and the second is the one that costs money. First: platforms you build on get sunset. If a workflow of yours — content moderation, data entry, QA sampling, survey panels — routes through one vendor's marketplace, you inherit their roadmap decisions. When they wind it down, you migrate on their clock, not yours.
Second, and bigger: "a human did it" is no longer proof of anything. When the cheapest labor on earth is quietly piping your task through ChatGPT, the label you paid for is a model's guess wearing a human's badge. That changes how you should design any pipeline that feeds a model or a report. Stop treating a data source as trustworthy because a person touched it. Build verification in: sample the outputs, cross-check against a gold set you control, and measure drift over time. The teams that get burned this year are the ones who assumed the input was clean because it came with a receipt.
Key takeaways
- Amazon Mechanical Turk stops accepting new customers July 30, 2026; existing accounts continue with no new features
- A 2023 study found 33–46% of MTurk workers used LLMs to do "human" tasks — undermining the data's reliability
- Platform sunsets are vendor risk: don't route a core workflow through a marketplace you don't control
- "A human did it" no longer means clean data — build sampling, gold sets, and drift checks into every pipeline
Feeding a model or a report with data you didn't verify? We build data and automation pipelines you actually own — with sampling, gold sets, and quality checks baked in, not bolted onto someone else's marketplace. See what we build or tell us what your workflow runs on today.
Sources: TechCrunch, Amazon Mechanical Turk.
- #mechanical-turk
- #data-quality
- #vendor-risk
- #ai-automation
- #data-pipeline
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
ByteDance, Alibaba kill AI companions — own your features
China's new anthropomorphic-AI rules force Doubao and Qwen to pull custom AI personas July 15 — a lesson in building AI features you don't control.
Read itAssort Health's $120M: voice AI runs the front desk
Assort Health raised $120M at $1.2B for voice AI agents that handle scheduling, intake, and refills. What it means for your phone line.
Read it