GPT-6 Astra targets junior-banker research and pitchbooks. Built with Morgan Stanley and Evercore, it runs a bank’s data through OpenAI’s servers.
OpenAI has launched ChatGPT for Financial Services, a version of its enterprise product built to do the company research, financial modelling and pitchbook work usually handled by junior investment bankers.
Morgan Stanley and Evercore worked with the AI company as design partners. The product runs on GPT-6 Astra, OpenAI’s newest model, and starts with investment banking and equity research teams.
Nick Turley, OpenAI’s Vice President of Product, told CNBC the tool was designed to research like an analyst and to back its conclusions like one, describing it as productivity rather than a replacement for bankers.
The launch comes as the AI company prepares for an expected initial public offering (IPO).
It says ChatGPT for Financial Services researches across multiple sources, traces figures across reporting periods and reads annotations in public filings.
Teams can build interactive charts and generate pitchbooks styled to a firm’s own template, and administrators can publish Excel, Word and PowerPoint templates through an admin page. OpenAI gave the example of a banker running a profit-and-loss normalisation to see which costs were excluded from an adjusted EBITDA figure before using it in a valuation.
Who keeps the data?
The product bundles datasets from Daloopa, PitchBook, LSEG News and Crunchbase, covering earnings transcripts, financial statements and company fundamentals.
The AI company indexes and hosts the data on its own infrastructure, as opposed to reaching providers through the MCP connectors its standard enterprise product uses. Existing third-party subscriptions can still be connected to ChatGPT for Financial Services.
Hosting the data on OpenAI’s servers routes a bank’s sensitive material, including deal research and client analysis, through a third-party AI layer.
The company said business content from ChatGPT Enterprise is not used to train its models by default, is encrypted at rest and in transit, and can be held under retention and data-residency controls, with zero data retention available on eligible endpoints for qualifying customers.
A 2025 preservation order in the New York Times copyright case required OpenAI to retain deleted and temporary ChatGPT logs, overriding its normal 30-day deletion schedule.
ChatGPT Enterprise accounts and customers with zero-data-retention agreements were excluded, though a court order terminated the going-forward obligation as of 26 September 2025, and the case continues. The company has published these terms for ChatGPT Enterprise but has not separately detailed them for the financial-services product.
The AI company said compliance teams can export workspace logs into audit workflows and that the product uses role-based access and encryption. Adoption beyond pilot programmes has not been demonstrated; the product essentially depends on banks trusting a third-party AI layer with confidential deal data.
OpenAI’s payments push
OpenAI is building into financial services from two ends. Its newly-launched investment-banking tool follows the Stripe and OpenAI launch of Instant Checkout in ChatGPT, and the released of an Agentic Commerce Protocol, an open standard that lets a shopper pay inside the chat without handing card details to OpenAI. PayPal, Visa and Checkout.com have adopted the protocol. Stripe’s Shared Payment Token initiates the payment without exposing the buyer’s card.
Synchrony agreed an enterprise collaboration with the AI company on 17 August to bring financing, rewards and loyalty into AI-native checkout, and expects integrating general-purpose cards into ChatGPT to take six to 12 months.
NatWest became the first UK bank to partner with OpenAI, in 2025. Who the paying agent legally is, and who is liable when it buys the wrong thing, remain unresolved.
