Rachel Sawyer, a technical writer based in Texas, has been involved in evaluating Google's AI products since March 2024. While her initial tasks involved verifying meeting notes and chat summaries generated by Gemini, her role gradually shifted toward censoring extreme content, including violence and sexual expressions.
Google AI Evaluation Workers Report Overwork and Low Pay
This article is a translation. Read the Japanese original
Sawyer stated that she was not asked to sign a consent form during onboarding, and the job description contained no mention of censorship work. Furthermore, she has suffered from anxiety and panic attacks due to the pressure of processing dozens of items daily under a strict time limit of 10 minutes per item. No mental health support was provided by the employer.
Sawyer is one of thousands of AI evaluators contracted with Google through Hitachi's GlobalLogic. The Guardian spoke with 10 current and former employees. Google also maintains contracts for AI evaluation services with Accenture and Appen.
While these workers earn more than data annotators in Nairobi or Bogotá, their pay is significantly lower than that of engineers in Mountain View. They occupy the middle tier of the global AI supply chain, performing the quality control necessary to ensure AI models provide safe and appropriate responses to users.
Source: ‘Overworked, underpaid’ humans train Google’s AI（HN 287pt・151コメント）（HN Search (backfill)、2025-09-13）