According to documents seen by Bloomberg, the effort - codenamed Project Mercury - involves ex-employees from major Wall Street firms such as JPMorgan Chase, Goldman Sachs, Morgan Stanley, KKR, and Evercore.

The recruits are reportedly paid $150 per hour to create and refine AI training data that mimics the work of junior analysts and associates, from IPO modelling to restructuring transactions.

Participants are supposedly required to build financial models in Microsoft Excel, adhering to strict industry standards, such as formatting margins and italicising percentages.

They also write detailed prompts explaining their modelling steps - guidance that will help the ChatGPT maker's systems learn how to automate such processes in the future.

Each contractor reportedly submits one model per week and receives feedback from reviewers before the data is integrated into OpenAI’s training pipelines.

The application process for Project Mercury itself is said to be AI-driven, with candidates first reportedly having to complete an interview with a chatbot based on their résumé, followed by a financial statement test and a final modelling assessment.

Many participants are said to be MBA students from Harvard and MIT, while others are seasoned analysts seeking flexible remote work.

An OpenAI spokesperson confirmed the company collaborates with experts “to improve and evaluate the capability of our models across different domains,” but said contractors are “recruited, managed and compensated by third-party suppliers”.

The supposed project marks a major step in OpenAI’s push to make its technology indispensable to industries such as finance, law, and consulting, where automation could drastically reduce repetitive work.

However, it also fuels anxiety among junior bankers, who already face gruelling 80-hour work weeks and now fear their roles could be the first casualties of the AI revolution.