OpenAI’s launch description says administrators can publish Excel, Word and PowerPoint templates and apply firm style guides. It also describes role‑based controls, configurable retention, supported compliance‑log exports and business data excluded from model training by default. These facilities can help a firm establish a consistent workflow, but a polished template does not demonstrate that its figures, assumptions or source permissions have been checked.
Workspace owners can enable financial‑data permissions through defaults or role overrides after account setup, according to the setup instructions. A useful first assignment would have a defined question, an approved set of sources and a named intended audience. For the hypothetical company screen, the handoff could include the screening criteria, why each company qualified, the dates of the financial inputs and a list of fields still missing. A reviewer can then assess the actual reasoning instead of guessing how a confident‑looking result was assembled.
OpenAI’s treatment of firm data is also separate from what the customer may do with licensed inputs. Under the partner terms, PitchBook forbids using its data for training, fine‑tuning, grounding or model development. Reuters restricts using its content, or associated data, signals or output, to train or develop AI models that persistently encode training data. A firm evaluating an internal model project therefore needs a separate permission assessment for its intended sources; the product’s own privacy setting does not settle that reuse question.
Photo: The Pioneer Building in San Francisco, photographed on July 27, 2019, when it housed OpenAI offices. Source: Wikimedia Commons. HaeB, CC BY‑SA 4.0. Previously cropped by the source; resized and cropped for display under the same licence.