Source: StoryBoard18

StoryBoard18 reports that OpenAI, in collaboration with training firm Handshake AI, is requesting third-party contractors to upload authentic files from their past and current employment. The goal of this data acquisition strategy is to train generative AI models for white-collar task automation using real-world examples such as Word documents, spreadsheets, and code repositories. Contractors are asked to describe professional tasks in detail and provide actual file outputs rather than summaries, allowing OpenAI to measure model performance against human baselines for complex professional work.

Legal experts have warned that this practice carries significant intellectual property and trade secret risks. While OpenAI provides a "Superstar Scrubbing" tool intended to remove personal and proprietary information, the responsibility for redaction rests entirely on the contractors. This raises questions about the potential for accidental leaks of confidential data belonging to former employers who did not consent to their materials being used for AI training. As OpenAI seeks specialized data to move beyond general capabilities toward domain expertise, this aggressive sourcing reflects the intensifying pressure to find high-quality training materials while navigating a complex landscape of data ethics and employment law.