← Capa | Mercado

OpenAI Hires Hundreds of Contractors to Manually Review ChatGPT Logs Under 'Project Lilly'

2026-11-04 · Fonte:Tom's Hardware

OpenAI Hires Hundreds of Contractors to Manually Review ChatGPT Logs Under 'Project Lilly'

A 404 Media investigation says OpenAI's 'Project Lilly' employs hundreds of contractors to read ChatGPT transcripts — including those containing personal information — to improve responses.

OpenAI has brought on hundreds of contractors whose job is to manually read and assess ChatGPT conversations, according to a report by 404 Media. The initiative, identified by the outlet as Project Lilly, involves human reviewers examining chatbot logs with the goal of refining the quality of the model's answers. Notably, the transcripts being reviewed include conversations in which users shared personal information, underscoring the breadth of access these contract workers have to real user interactions.

The report, which Tom's Hardware also covered, casts light on a largely unseen layer of the AI production pipeline: the human workforce that grades, inspects and troubleshoots model outputs. While much of the industry's attention centers on chips, clusters and training budgets, Project Lilly demonstrates that conversation-level review by people remains central to how a flagship chatbot is improved. The scale — hundreds of contractors — signals a substantial, ongoing operation rather than occasional spot-checking.

For enterprises weighing ChatGPT deployments, the findings carry practical weight. Human review of transcripts — including those containing personal information — means prompts submitted to the assistant may be read by contractors, a consideration for organizations handling sensitive data or operating under strict compliance regimes. 404 Media frames the practice as part of OpenAI's effort to improve responses, but the disclosure is certain to draw scrutiny from privacy-conscious enterprise customers evaluating how their interactions are processed.

The story also illustrates the labor economics behind AI products. Alongside compute budgets sit workforces hired to evaluate and grade model outputs — work that touches conversations containing personal information. As vendors compete on response quality, reports like this one suggest the competition extends beyond hardware: it runs through a contractor workforce reading user logs, a reminder that frontier AI operations depend on people as much as silicon.

Tags: OpenAI · ChatGPT · Project Lilly · Human Review · Data Privacy


Leituras relacionadas


萬安算交所 AIXX · AIXX · Exchange verificado · Ir ao mercado