OpenAI is reportedly asking third-party contractors to upload examples of real work produced in past and current jobs as part of an effort to generate higher-quality training data for its AI models, according to a report by Wired. The initiative, carried out with training data provider Handshake AI, reflects a broader industry push to use authentic professional outputs to improve models designed to automate white-collar tasks.
Materials reviewed by Wired indicate that contractors are asked to describe job tasks and submit concrete work products, such as documents, presentations, spreadsheets, images, or code repositories. OpenAI reportedly instructs contributors to remove proprietary and personally identifiable information and provides tools to assist with data sanitization.
Legal experts have cautioned that the approach introduces potential intellectual property risk. Evan Brown, an intellectual property lawyer, said the model relies heavily on contractors’ judgment in determining what information is confidential, creating exposure for AI developers as they scale training operations.
