Introducing ChatGPT for Financial Services

OpenAI is introducing ChatGPT for Financial Services: a tailored ChatGPT Work experience that pairs built-in financial data with GPT‑6 Astra’s reasoning for research, financial models, and client materials. Design partners Morgan Stanley and Evercore helped aim the first cut at investment banking and equity research — where reliable data access and high-quality artifact creation were the loudest pain points.

Premium data from providers such as Daloopa, PitchBook, LSEG News, and Crunchbase is indexed and hosted by OpenAI, with granular citations so figures and claims can be traced to sources. Firms also get central control of access and data connections on top of ChatGPT’s enterprise security and governance controls. Availability is limited to eligible financial institutions.

We’re watching OpenAI productize the banker desktop. The hard problem in this vertical is not chat fluency — it’s trusted numbers, entitlements, and artifacts that survive a compliance review. Hosting Daloopa, PitchBook, LSEG News, and Crunchbase inside OpenAI’s infrastructure is how they claim better retrieval, latency, and citations without every desk wrestling MCP connectors. They’re also working with S&P Capital IQ, LSEG, MSCI, Dow Jones Factiva, and Moody’s on shared sign-in so existing subscriptions can flow through ChatGPT entitlements, plus a broader connector set that OpenAI says includes more than 50 providers.

GPT‑6 Astra is positioned as strong on information retrieval, financial reasoning, and artifact generation — documents, spreadsheets, slides — with admins able to publish firm Excel, Word, and PowerPoint templates. On OfficeQA Pro, which tests agents on U.S. Treasury Bulletins including complex tables and footnotes, OpenAI reports GPT‑6 Astra at 69.9% versus 60.2% for GPT‑5.6 Sol. That’s one public benchmark on the page; we’re not treating it as a banking-quality bar.

Our read: this is useful AI if citations are real and MNPI stays behind the enterprise wall. OpenAI says business data is not used to train models by default, is encrypted at rest and in transit, and supports SAML SSO, SCIM, RBAC, retention controls, compliance log export, and multiple workspaces for information barriers. The downside if it scales is concentration of research workflow — and liability — inside one frontier vendor’s hosted data plane. A wrong EBITDA reconciliation that looks cited is worse than a blank cell. I’m watching whether design-partner methods become the product’s post-training diet, and whether banks treat Astra output as a draft that must clear a human, or as advice with a logo.

OpenAI frames this as one path into finance alongside the API for firms that build their own apps. Provider quotes on the page stress private-market coverage (Crunchbase, PitchBook) and fundamental data for agentic workflows (Daloopa). Expansion beyond banking and equity research is described as forthcoming, informed by partner work.