OpenAI faces scrutiny over use of real job files for model training

OpenAI is reportedly asking third-party contractors to upload examples of real work they have completed in previous and current jobs.

According to Wired, the request is part of a broader trend across the artificial intelligence (AI) industry, where companies rely on human contractors to generate high-quality training data. 

The goal is to improve AI systems so they can eventually automate more white-collar and knowledge-based tasks.

As per the report, OpenAI, working alongside training data firm Handshake AI, has asked contractors to describe tasks they performed at other workplaces and upload examples of actual on-the-job outputs. 

These materials may include Word documents, PDFs, PowerPoint presentations, spreadsheets, images, or even code repositories, rather than summaries or recreated samples.

OpenAI has instructed contractors to remove proprietary information and personal data before submitting any files. 

Contractors are also directed to use a ChatGPT-based “scrubbing” tool designed to help eliminate sensitive content from documents.

Despite these safeguards, legal experts have raised concerns. Intellectual property lawyer Evan Brown told Wired that this approach carries significant legal risks. 

He warned that it relies heavily on contractors correctly judging what information is confidential, which could expose AI companies to intellectual property disputes.

The report highlights ongoing tensions between AI development needs and data ownership concerns.