OpenAI released on September 10 a version of ChatGPT built for finance, and it recruited two of the firms that will compete with it to help design the thing. ChatGPT for Financial Services is a finance-focused edition of the company’s ChatGPT Work product, with Morgan Stanley and Evercore as design partners.
The underlying model is GPT-6 Astra, but the product is defined less by the model than by the data it can reach. The service connects to LSEG, Daloopa, and PitchBook, which means it can pull filings, earnings-call transcripts, and company data directly, then turn them into research and financial analysis.
The output is meant to look like what a junior banker produces. The tool does company research, runs financial analysis, and builds presentation materials, and it carries the three features a regulated firm will actually check for: citations that trace each claim to a source, verification of charts, and permission controls for confidential deal material.
Nick Turley, the product lead, framed the ambition in plain terms at the launch briefing. OpenAI is teaching ChatGPT to do research the way an analyst does, and to back its conclusions the way an analyst would, with a trail of sources rather than an assertion.
The initial focus is narrow on purpose. The product is aimed first at investment banking and equity research, the parts of finance where the work is structured, the data is standardized, and the value of a faster junior is easiest to measure.
The pitchbook is the telling example. Building a pitchbook, compiling comparable companies, and checking figures against filings is the daily work of first-year analysts and associates, and it is also the on-ramp of the banking apprenticeship. A tool that does that work competently does not just save hours; it changes who a firm needs to hire.
OpenAI is not first to the market. Anthropic released Claude for Financial Services a year earlier, aimed at the same desks, and the two companies are now competing for the same budgets with products that differ mostly in whose model and whose safeguards a bank trusts.
The design-partner model is the notable choice. Morgan Stanley and Evercore are not neutral customers; they are firms whose own analysts the product could eventually displace, and enlisting them as designers is a way for OpenAI to buy credibility while the tools are still being shaped.
Analysts said the appeal to banks is straightforward. The industry runs on documents and models, and the cost of a first-year analyst is high relative to what the work produces. A product that compresses that cost is an easy sell to a business that has spent a decade cutting headcount in the name of efficiency.
The permissions layer is what gets it past compliance. Deal work is confidential by law, and no bank will run its M&A material through a tool that might train on it or leak it. OpenAI’s answer is controls that keep confidential material inside the customer’s perimeter, which is the feature the design partners presumably spent the most time testing.
The deeper question is what happens to the apprenticeship. If the compiling and checking work is automated, the junior roles that taught the next generation of bankers thin out, and the industry is left with fewer people who learned the trade by doing the grunt work.
For OpenAI, finance is a beachhead into the enterprise accounts that pay the largest software bills. A model that proves itself on Wall Street, where the stakes are dollars and the errors are expensive, is a model every other industry will take seriously.
Morgan Stanley is not a new partner. The bank was an early adopter of OpenAI’s tools, rolling out a GPT-4-based assistant for its wealth-management advisors in 2023, one of the first large-scale deployments in finance. Its return as a design partner suggests the relationship has deepened from pilot to product.
The data partners carry their own weight. LSEG, the owner of the London Stock Exchange and a major data vendor, has its own partnership with Microsoft to embed AI across its terminals, and its inclusion gives the product a direct line to the market data banks already pay for.
The sell-side economics are the real driver. Equity research has been shrinking for a decade as passive investing and fee compression squeezed the business, and analyst headcount at banks has fallen accordingly. A tool that reproduces junior research work at a fraction of the cost is aimed at an industry that has already decided to do more with fewer people.
The competitive edge is not capability but trust. The models are close enough that the decision will turn on data access, security, and whether the output can be defended in front of a client, and that is the ground OpenAI is now trying to claim.
