On September 21 @ChatGPT posted that you can track your credit score in the product. Connect Experian. Get a VantageScore 3.0. Stay informed when things change. The same day’s help pages added a Privacy Center. OpenAI’s Data controls article now points to it. The Finances help page is the source for the bureau hookup. The Privacy Center page is the source for the menu.
I wrote on the 16th that Project Lily is contractors grading real chats, and that the training toggle does not unread them. Today’s product is the same box, with a credit file in it, and a new page that explains the old switches.
What they shipped
Finances is already in the U.S. for Plus and Pro on web, iOS, and Android. Banks go through Plaid. Credit is a second pipe. You do not need a bank connection to use the score. You fill Experian’s identity form, not the chat. OpenAI says it receives your connection status and the credit data you authorize Experian to share. Checking the score this way is a soft inquiry.
What you can see is the Experian report and a VantageScore 3.0 from 300 to 850. You can ask about reported accounts, balances, credit limits, utilization, payment history, and inquiries. The score and report update monthly. Credit monitoring can alert you to a new inquiry, a new account, or a new address. You can turn alerts off and keep the connection.
The leftover is in the help page. Disconnect Experian and the bureau stops sharing with ChatGPT. OpenAI deletes the underlying report data from its systems within 30 days. Disconnecting does not remove information from past conversations or financial memories. Those you delete separately. Conversations with Finances follow the same Improve the model for everyone setting as the rest of ChatGPT.
There is a third copy you type yourself. Financial memories. A mortgage. A loan from your parents. Rent split with a partner. How much you want in checking. ChatGPT can save that from a finance chat and use it later. Temporary chats do not touch the connected accounts or those memories. A regular chat does.
The Privacy Center is rolling out to signed-in Free, Go, Plus, Pro, and Business. Not Enterprise, Edu, or Healthcare. On the web it lives under Help. Opening it does not change any setting, delete any data, or override a workspace policy. It has three folders. Personalized chats and ads: Memory, personalized ads, location. Chat privacy: temporary chat. Data use and access: plugins and apps, MFA, model improvement, export or delete. Ads may appear for eligible Free and Go users. Paid work plans do not get them.
The copy on the hub says OpenAI does not sell your data and does not share ChatGPT conversations with advertisers or marketing partners. The same hub has a control for personalized ads. Those are different sentences. Both can be true. Neither is a fence around a credit file.
What the hub does not change
The training toggle is still in Data controls. The Privacy Center will walk you to it. It will not flip it. Free, Plus, and Pro still start with Improve the model for everyone on. Turning it off applies to new chats. It does not pull back what already left. It does not stop authorized staff and vendors from seeing content for abuse, support, or legal work.
If the toggle is on, a Finances conversation is a conversation. Ask which card is driving utilization and the answer cites the report. That thread is eligible for model improvement the same way a recipe is. Project Lily was style work on real prompts. The Privacy Filter is still a model. The reviewer is still a person. A credit factor in the prompt is not a special class in the help page.
Thirty days is the other number. Temporary chats may be retained that long for safety. Synced Experian data sits that long after you disconnect. Plaid data too, on both OpenAI and Plaid, according to the Finances article. A hub that explains the 30 days is not the same as a product that does not keep the file.
VantageScore 3.0 is not the score every lender uses. OpenAI’s own disclosure says that. The product still invites you to ask about a mortgage from inside the same session that holds the report. The session is the thing OpenAI already said it retains until you delete it.
I do not need a theory of alignment for this. If you can ask the bot about your utilization, the report is in the session.
Keep the file off their disk
The useful control is not a new Help item. It is whether the credit report is a file you already have at Experian, and whether a chat about it ever exists.
If you want the score, Experian’s own site still has it. If you want a bot that has seen your utilization, you are asking the same product that still defaults to training, still has a contractor pipeline, and still keeps chats after you unplug the bureau. Local models still do not give you Astra. They also do not ask for your Experian login.
The hub is a map. The score is another copy.