OpenAI introduced ChatGPT for Financial Services on September 10, 2026. The product combines a tailored ChatGPT Work experience with built-in financial data and GPT-6 Astra reasoning. The important change is not simply a stronger model: OpenAI is packaging data access, citations and financial workflows into a product designed for institutional users.
What OpenAI says the product includes
OpenAI says the financial-services experience combines built-in financial data with GPT-6 Astra to help teams develop research, financial models and customized client materials. The announcement names Daloopa, PitchBook and LSEG News as data providers. OpenAI says the data is indexed and hosted by OpenAI rather than requiring users to connect those sources through separate MCP connectors.
Why built-in data changes the workflow
Traditional AI workflows often require a user to find documents, connect a data source and then ask a model to analyze the result. OpenAI's design puts the data layer inside the product. The company says this enables granular citations so bankers can trace figures and claims back to sources while analysis is being developed.
That distinction matters for research-heavy work. A citation attached to a generated claim is useful only if the underlying source is available and relevant. OpenAI's announcement describes the product direction, but it does not provide enough public detail to independently assess every data refresh interval, coverage rule or citation ranking behavior.
How the Morgan Stanley and Evercore work fits in
OpenAI says the product was shaped by design partnerships with Morgan Stanley and Evercore. Those partnerships are evidence of product development with financial institutions, not proof that every bank will use the same workflow. The announcement says the work helped identify where OpenAI could solve challenges for financial institutions and informed the solutions being built.
What financial teams should verify before using it
Teams should separate three questions: whether the required financial source is covered, whether the data is current enough for the task, and whether the generated analysis can be reviewed against the cited evidence. A model can reason over a source without that source being suitable for every financial decision.
The product is also different from simply giving a general-purpose model access to a spreadsheet. Its value proposition depends on the combination of specialized data, the Work interface and traceable citations. If any one of those layers is weak for a particular use case, the overall workflow may need additional systems.
What is public and what is still unclear
OpenAI has publicly described the named providers, the GPT-6 Astra model, the Work experience and the citation approach. The announcement does not fully specify all supported financial workflows, detailed entitlements, every dataset field, model quotas or the exact behavior of every financial-data query. Those should be confirmed with OpenAI before a production rollout.
The practical takeaway for financial-services teams
The strongest use case is evidence-heavy work where analysts need both synthesis and a path back to source material. Research, financial-model preparation and client-material drafting are named by OpenAI itself. The sensible operating model is still review-first: use the system to accelerate analysis, then verify material figures and claims against the cited source before they enter a client deliverable.
Three useful evaluation questions
Ask whether the required datasets are covered, whether citations let an analyst verify each important figure, and whether the product fits the firm's existing review and data-governance process. Those questions test the actual product rather than assuming that a financial-specific AI label automatically makes an output reliable.