On Thursday, September 10, OpenAI unveiled ChatGPT for Financial Services, a version of ChatGPT Work built for the finance industry and powered by GPT-6 Astra, the company's most advanced model. The first phase covers two areas: investment banking and equity research. The product was not built in a lab: OpenAI developed it with Morgan Stanley and the advisory firm Evercore, both listed as design partners that helped shape which features belong in the product and how they should behave. It is not a generic assistant with a few extra finance instructions, but a vertical product that starts from the data formats and workflows banks and research teams already use.
"We're effectively teaching ChatGPT to do research like an analyst and to defend that research like an analyst," said Nick Turley, vice president and head of ChatGPT, in a briefing with reporters. The stated goal is to cut the manual work that consumes most of an analyst's day: gathering data, rebuilding models, formatting materials for clients. Turley said he had spent time in the days before launch at Morgan Stanley's and Evercore's offices working alongside analysts to see where the model got things wrong and where it genuinely sped up an existing process.
The main difference from the standard enterprise edition is in the data. ChatGPT for Financial Services ships with pre-loaded datasets from LSEG News, Daloopa, PitchBook and Crunchbase: earnings transcripts, financial statements, company fundamentals and information on private companies. Firms with existing Bloomberg or FactSet subscriptions can connect them and use the data they are entitled to without changing vendor, while the MCP connector ecosystem reaches about fifty integrations. When the model works from that data it produces citations that trace back to the source: every figure can be linked to the original document and charts can be audited against the underlying numbers. For a regulated institution that is the difference between a plausible output and a traceable one, which is the difference between a demo and a process that can be approved.
In the demonstration shown to reporters, the system analysed a potential merger target: it selected the comparable set, pulled market data and built a deck in PowerPoint using the bank's template. The hard part, Turley said, was not slides that look good but slides that make sense: the model had to pick the right peers, load prices into a spreadsheet, verify the chart against the data and explain the stock's sell-off. The product also generates Excel workbooks and web dashboards, and lets users set the model's reasoning effort across three levels: more effort means more tokens consumed, a higher cost and, according to OpenAI, a better output. Administrators can pre-load company templates: "A new intern on day one has all this ready to go," said Joseph Kim, the product lead.
Access is not open. Firms need an enterprise account and must qualify as eligible institutions under criteria OpenAI has not published: the company asks to be contacted directly. The product rests on ChatGPT Enterprise controls, with SAML single sign-on, SCIM provisioning, role-based permissions, encryption and retention windows that administrators can configure, plus specific controls for sensitive deal materials. OpenAI said customer data is not used to train its models by default.
The launch targets precisely the work banks have historically handed to new graduates: research, models, pitch books. Turley rejects the idea of job cuts and insists on productivity: "Analysts and bankers work 100-hour weeks," he said, and the effect will resemble that of Excel, which let bankers produce better analysis faster. The parallel has a limit, though, because Excel executed the analyst's reasoning faster, while here the model chooses the peers and explains the sell-off.
The concern has come from inside the industry too. Chris Churchman, a Goldman Sachs partner who helps lead artificial intelligence efforts in Global Banking and Markets, has warned of cognitive atrophy for the next generation of financiers if reasoning is outsourced to models. The labour-market data points the same way: according to Stanford Digital Economy Lab research updated in August, employment for 22- to 25-year-olds in occupations most exposed to generative AI is 19% below where it would have been had they followed less-exposed peers, a decline attributed to fewer hires rather than layoffs. The Dallas Fed reached similar conclusions in January. If a model produces a defensible pitch book in minutes, the economics of the two-year analyst programme change: the role does not vanish, it shifts from assembly toward verification.
OpenAI is arriving after its rivals. Anthropic launched its own finance offering back in 2025, and narrower workflows are already served by startups such as Shortcut and Endex. The prize, however, is larger than one product. OpenAI comes to Wall Street with more than a billion weekly ChatGPT users and 2.5 million businesses using its tools, and in August chief financial officer Sarah Friar told investors that the enterprise business had overtaken consumer as a source of revenue. In June the company launched a Partner Network with $150 million committed and a goal of certifying 300,000 consultants by the end of the year. Turley has said vertical products for other industries will follow, with the stated ambition of becoming the canonical product for banks with tens of thousands of employees.
Holger Mueller, an analyst at Constellation Research, summed up the challenge: OpenAI has done the easy part and must now prove security and real value in regulated verticals. That, more than any demonstration, will decide whether ChatGPT takes a permanent seat in financial decision-making or remains a document accelerator.
Sources
