OpenAI builds eleven licensed data providers into its finance ChatGPT

The second-year analyst building a leveraged buyout model at 11pm already has a market data terminal open in one tab and a private-markets database in another. OpenAI's version puts those feeds behind a prompt in a single workspace ^[1]. That is who this is for: the person assembling the deck, not the portfolio manager reading it.

Eleven providers are named ^[13]. The negotiation behind that list is the expensive part, because permission to use a vendor's data inside another company's product can take time and involve complicated licensing agreements ^[11]. TNW's read is that the data deals may be more important than the model itself ^[12].

What a buyer cannot do yet is the arithmetic. OpenAI has not published a price ^[3]. The reporting also does not say whether the licensed feeds appear for every buyer or only for firms that already subscribe to them ^[24]. Those two gaps decide whether this is a cheaper front end on data a bank already pays for, or a second bill for the same content.

On the model, the gain on OpenAI's OfficeQA Pro test of financial document analysis is 9.7 percentage points ^[14]. The figure that governs rollout is the other one: 30.1 percent of the test still comes back wrong ^[15]. TNW notes that an error there can end up in a valuation model or a client presentation ^[8].

The controls read like a list written by someone who has sat through a bank vendor review: single sign-on, user provisioning, role-based access, no training on business data by default, encryption at rest and in transit, configurable retention and audit log exports ^[16]. Information barriers run as separate workspaces, which TNW describes as a practical way of dealing with the Chinese walls investment banks use to keep sensitive information apart ^[17].

The European picture is thinner. Under DORA the obligations sit with the financial entity: a register of ICT providers, contract terms covering subcontracting and audits, an exit strategy and resilience testing ^[20]. TNW reports that many European financial firms are still not ready for those requirements ^[21]. DORA also lets European supervisors designate a technology provider as a critical ICT third-party service provider, and designated firms face direct EU oversight including inspections and recommendations that institutions may have to act on ^[22]. No AI model provider has been designated under that mechanism ^[23].

Morgan Stanley helped design the product ^[2]. Morgan Stanley analysts have also predicted that European banks could lose a fifth of their jobs to AI ^[9]. The model underneath was released on 3 September with OpenAI's own warning about its cyber capabilities, and ENISA was given access this week to test it ^[10].

For anyone rolling this out, the decision splits on two answers. The first is whether a feed appears only for firms that already license it; if it does, the comparison is against the seats you can drop, and if it does not, you are paying twice for the same content until a renewal. The second is where the output lands; internal screening and first-pass research can absorb a 30.1 percent miss rate with a spot check, and a client-facing document needs a named reviewer and the hours that costs ^[15]. A European entity doing client-facing work has the hardest cell of the four, because its register entry and exit plan are due under a regime the announcement does not address ^[18].