OpenAI hires ex-bankers to train AI for finance

OpenAI is using over 100 former investment bankers to train its AI to build financial models, aiming to automate tasks typically handled by junior bankers.

The initiative, known as Project Mercury, involves ex-employees from firms such as JP Morgan Chase, Morgan Stanley, and Goldman Sachs, according to documents.

Participants are paid US$150 per hour to develop prompts and create models for transactions like restructurings and IPOs, and are given early access to the AI tools under development.

OpenAI’s spokesperson said the company works with various experts to improve its models, who are hired and managed by outside suppliers.

🔗 Source: Bloomberg

🧠 Food for thought

Implications, context, and why it matters.

U.S. regulators have not issued AI-specific rules for live-deal modeling, creating uncertainty

OCC labeled AI an emerging risk in December 2023 and said banks must apply the same risk standards as with other tech, yet rules do not address AI 1.
Acting comptroller Michael Hsu cited unchecked growth in derivatives plus cryptocurrency before crises and urged checkpoints before AI moves from “input provider” to “co-pilot” to “autonomous agent” (systems that can act without direct human oversight) 1.
OCC set a December 2024 deadline with a June 2025 presentation, signaling frameworks are in progress 2.
FINOS launched Common Controls for AI Services in June 2025 with major banks and cloud providers to set controls for AI use, which suggests industry-wide standards do not yet exist 3.

Banks plus AI vendors add third-party governance for model risk and audit needs

OpenAI’s Mercury hires contractors to build Excel models for live deals using “industry norms for formatting”. It does not mention audit trails or data controls or formal model risk documentation.
Member firms in FINOS said fragmented approaches are insufficient for AI in regulated markets and are building tech-neutral controls that span banks, cloud providers, plus AI vendors 3.
Industry leaders call out two success factors for AI use, risk-proportionate governance and human-in-the-loop design, which keep people responsible for review plus approval 4.
Napier AI sells anti-money laundering compliance software, while Ocrolus offers document automation and data extraction for lenders; both already serve banks with AI built for compliance 5.