OpenAI introduced ChatGPT Financial Services on September 10, 2026, as a tailored ChatGPT Work experience for financial institutions. It combines built-in financial data with GPT-6 Astra reasoning for research, financial models, and client materials. The point is not simply to give analysts a chatbot with more finance knowledge, but to place retrieval, analysis, production, and review in one workflow.
OpenAI says the product was shaped with design partners including Morgan Stanley and Evercore, initially focusing on investment banking and equity research. These are vendor-reported early examples rather than independently reproduced effectiveness studies. They do show where OpenAI is directing the product: work that previously required several data terminals, spreadsheets, and presentation tools to operate together.
The built-in premium data providers include Daloopa, PitchBook, LSEG News, and Crunchbase. OpenAI says the data is indexed and hosted by OpenAI, with citations that can point to specific tables or passages. For financial research, that detail matters more than a general claim that the model knows more. An analyst needs to know which source, field, and passage support a number to reproduce the work and handle version changes.
The product can also work with existing subscriptions through services such as S&P Capital IQ, LSEG, MSCI, Dow Jones Factiva, and Moody’s. OpenAI describes shared sign-in and entitlement integrations that use access a firm already holds. That can preserve parts of an existing procurement and access model, but deployment teams still need to check vendor terms, account tiers, data residency, and the read/write scope of each connector.
GPT-6 Astra is positioned here as more than a question-answering model. OpenAI says it can handle retrieval and financial reasoning, navigate figures, tables, and notes, and create documents, spreadsheets, presentations, and interactive charts. Keeping citations to tables and passages in the output makes human review closer to checking an evidence chain than guessing whether a model misread a page.
Administrators can publish Excel, Word, and PowerPoint templates as well as firm style guides so that research notes, valuation models, and pitchbooks follow an institution’s format. OpenAI also describes a connector ecosystem of more than 50 integrations, with examples such as Datasite, Box, Preqin, FactSet, and Intapp, plus performance work for MCP connectors. As the tool count grows, permission design and data classification matter more than simply enabling more connectors.
The governance layer builds on Enterprise SAML SSO, SCIM, and RBAC, with encryption at rest and in transit, configurable retention, exportable compliance logs, and role-based control of skills and apps. Firms can enable or disable supported read/write actions and use multiple workspaces to create information barriers. OpenAI says business data is not used for training by default; production teams should still verify the contract, admin configuration, and actual behavior.
The broader change is a move from a general assistant toward a professional workflow with data rights, citations, templates, and audit requirements. Financial institutions still need to define data semantics, model boundaries, human judgment, and accountability, and eligible institutions need to contact OpenAI’s account team for access. The important question is not whether the product replaces analysts, but whether it shortens the path from trusted data to a reviewable deliverable.
