SAN FRANCISCO — OpenAI on September 10 unveiled ChatGPT for Financial Services, the first industry-tailored edition of its enterprise product ChatGPT Work, aimed squarely at the work Wall Street has assigned to junior bankers for decades: researching companies, dissecting financial data and building pitchbooks.
The timing is not hard to read. Chief financial officer Sarah Friar told investors in August that OpenAI's enterprise business now generates more revenue than its consumer business, and the company is widely expected to be preparing a blockbuster IPO. In the enterprise market it is fighting Anthropic — which launched Claude for Financial Services last year — and Google.
[2][1]What is inside
According to OpenAI's announcement, the product runs on GPT-6 Astra, the model that launched September 3, and ships with built-in paid financial datasets from Daloopa, PitchBook, LSEG News and Crunchbase, covering financial statements, earnings-call transcripts, company fundamentals and private-company information. OpenAI indexes and hosts the data itself, which it says improves retrieval accuracy and enables line-item sourcing: every figure and claim can be traced back to the original table or paragraph.
For institutions that already pay for data, OpenAI is working with S&P Capital IQ, LSEG, MSCI, Dow Jones Factiva and Moody's on unified login and permissioning, so ChatGPT recognizes the user and only surfaces data the firm has already licensed. The company also says it has hardened the MCP connectors most used in finance (S&P Global, FactSet) and that its connector ecosystem covers more than 50 tools.
The announcement cites one benchmark: on OfficeQA Pro, which tests whether agents can find and analyze information in the complex financial tables of U.S. Treasury bulletins, GPT-6 Astra scores 69.9 percent, versus 60.2 percent for the previous GPT-5.6 Sol; OpenAI's own comparison chart also lists Anthropic's Claude Fable 5.1 at 62.4 percent. These are self-reported numbers on a benchmark OpenAI chose; no third party has reproduced them.
[1]The demo and the pitch
At a briefing, OpenAI vice president of product Nick Turley demonstrated the product analyzing a potential M&A target: pulling financials from industry-standard data sources, selecting comparable companies, dropping prices into a spreadsheet, checking charts against the data, and generating a PowerPoint deck in the bank's own template.
"It's very easy to make slides that look good, but it's much harder to make slides [that] actually make sense," Turley said. "To get here, ChatGPT had to choose the relevant peers. It had to pull the prices into a spreadsheet. It had to check the chart against the data, and it had to explain the sell-off and the rebound."
[2]We're effectively teaching ChatGPT to research like an analyst and back up its conclusions like an analyst as well.
What happens to the junior bankers
Asked directly by CNBC whether the product would reduce banks' need to hire junior staff, Turley did not answer head-on. He offered the efficiency story instead: analysts work 100-hour weeks, and "in the same way that Microsoft Excel transformed the industry and allowed them to produce better analysis faster, you will see technology like this do the same."
The Excel analogy is clever, and incomplete. Last month Chris Churchman, the Goldman Sachs partner in charge of one of the bank's flagship AI projects, warned of the other side: automating the tasks that train junior bankers risks "cognitive atrophy" in the next generation of financiers. "Reasoning is still important," he said. "You still need to reason about [problems] and structure it into an argument, and now we're delegating reasoning."
The apprenticeship model depends on juniors building judgment through the repetitive labor of copying numbers, building models and revising decks. If the first draft always comes from the machine, the organization gains speed and the trainee loses the chance to be wrong. The launch did not answer that question. It may be the product's real price.
[2]Governance, availability — and what went unsaid
On security and compliance, OpenAI says the product inherits enterprise-grade controls: SAML single sign-on, SCIM provisioning and role-based access; business data is not used for training by default; encryption in transit and at rest; configurable retention; workspace logs that compliance teams can export into existing audit processes; and multiple workspaces for information barriers. These are table stakes for enterprise procurement, not regulatory certifications — the public materials contain no FINRA or SEC approval, no data-residency list and no SLA detail.
The product is available to eligible financial institutions through the sales team, with no public pricing. The initial focus is investment banking and equity research. Turley described demand as "a ton," but declined to name signed banks — Morgan Stanley and Evercore are "design partners" that helped shape the product, which is not the same as a public customer list. OpenAI also plans tailored versions for "a number of sectors" beyond finance.
Sourcing citations solve traceability, not correctness. A model can cite a real filing and misread the fiscal period; it can pull the right number and pick the wrong peer. What this launch proves is that OpenAI can package data licensing, citation chains and enterprise permissions into a product. What it cannot yet prove is how stable that product is on somebody else's deal materials.
[1][2]