Infra · 3 Jul 2026 · 2 min read

Amazon Stops Accepting New Mechanical Turk Customers

Amazon is closing Mechanical Turk to new customers, forcing builders to find more reliable platforms for human-in-the-loop data annotation.

Pen-and-ink illustration: a grand archway, its ke. For the story "Amazon Stops Accepting New Mechanical Turk Customers".
— Pen-and-ink illustration: a grand archway, its ke. For the story "Amazon Stops Accepting New Mechanical Turk Customers". —

What happened

Amazon will close its Mechanical Turk crowdsourcing service to new customers on July 30, 2026. The announcement, first reported by TechCrunch AI, confirms the platform is entering a maintenance-only mode. Existing customers can continue to use the service as normal. Amazon Web Services stated it will continue to invest in security and availability but has no plans to introduce new features.

How the room's reading it

The consensus among practitioners is that this was a long time coming. On forums like Reddit, users suggest the platform died 'years ago' from bots and fraud, making this announcement feel more like a formality than a shock. Data scientists have long worried about the platform's data quality, especially after a 2023 analysis found a high percentage of workers were using LLMs to complete their tasks. This poisoned the well for anyone relying on MTurk for genuine human annotation. The move is seen less as Amazon killing a product and more as the company finally acknowledging that the service is on life support.

Sailfish's take

We stopped using Mechanical Turk for serious data work years ago. The signal-to-noise ratio became untenable, and the irony of paying humans to run LLM outputs was too much. This isn't just about finding a replacement crowdsourcing platform — it's about rethinking the need for one. For many annotation and evaluation tasks, a well-prompted model often provides cleaner data than a disengaged human worker. We've found that a small, trusted set of internal evaluators combined with model-based checks gives us far more reliable results. If you're still relying on anonymous crowdsourcing for your core data loop, this is the week to stop.

Our take — your read?

Be the first to weigh in.

Sources
— END OF DISPATCH — Infra