New Delhi: OpenAI is shifting its focus closer to the workplace as it looks to make its next generation of AI models more useful for real-world tasks. The ChatGPT maker has partnered with training data firm Handshake AI to collect examples of routine professional work completed by third-party contractors, according to a report by Wired.
The initiative involves gathering data drawn from the actual responsibilities contractors handled in their current or previous jobs. The goal is to measure how OpenAI’s models perform against a human benchmark across a wide range of practical tasks, especially those common in enterprise settings.
This approach reflects a broader trend across the AI industry. Companies such as Anthropic and Google are also relying on large contractor networks to generate specialised, high-quality training data as they build AI systems and agents designed to automate office and knowledge work.
The move comes against a backdrop of growing anxiety about AI’s impact on white-collar employment. Several technology leaders have warned that automation could significantly disrupt entry-level and routine roles, even as companies like OpenAI continue to pursue artificial general intelligence, a form of AI that could outperform humans across many economically valuable activities.
Under the programme, contractors are required to submit real workplace tasks in two parts. The first is the original request from a manager or colleague, and the second is the completed output delivered in response. OpenAI has asked for concrete work products rather than summaries, including documents, presentations, spreadsheets, images, or code repositories.
In an internal presentation, the company reportedly encouraged contractors to upload examples of genuine on-the-job work they have already completed, specifying that submissions should be “a concrete output (not a summary of the file, but the actual file), e.g., Word doc, PDF, Powerpoint, Excel, image, repo.”
To address privacy and confidentiality concerns, OpenAI has instructed contractors to remove sensitive material before uploading files. This includes proprietary business information and personal data, using a dedicated internal tool designed for this purpose.
“We’ve hired folks across occupations to help collect real-world tasks modeled off those you’ve done in your full-time jobs, so we can measure how well AI models perform on those tasks. Take existing pieces of long-term or complex work (hours or days+) that you’ve done in your occupation and turn each into a task,” OpenAI said in an internal document cited in the report.
“Remove or anonymise any: personal information, proprietary or confidential data, material nonpublic information (e.g., internal strategy, unreleased product details),” it added.
The effort highlights the rise of a fast-growing ecosystem of data contracting firms, including Handshake AI, Surge, Mercor and Scale AI. These companies recruit and manage large pools of workers whose role is to supply the detailed, task-specific data now seen as essential for training more capable and commercially viable AI systems.
BI Bureau
