Is it safe for a regulated lender to run loan documents through Claude or ChatGPT?
It can be, on the right account with the right checks. What to confirm about the AI vendor, what never goes in, and how commercial and consumer lenders differ.
A regulated lender, including a commercial real estate lender, can run loan documents through Claude or ChatGPT safely only on a business account whose terms exclude training on your data, approved like any other vendor, with a contract that covers safeguards, and with a person checking every output before anyone relies on it. Free and personal accounts are the wrong place for a loan file.
Does the AI vendor train on what you upload?
It depends on the account. Anthropic says it does not use inputs or outputs from its commercial products (Claude for Work, Team, Enterprise and the API) to train its models by default, except where a user sends feedback such as a thumbs up or down, which an organization’s owners can turn off (Anthropic’s privacy centre). Consumer accounts follow different terms. Check the current terms of every tool you approve, and how long it keeps data. The provider checklist.
Is an AI vendor a service provider?
Treat it as one. In the US, non-bank lenders that handle consumer information are covered by the FTC’s Safeguards Rule, which requires firms to select capable service providers, require safeguards by contract, and periodically assess them. Banks apply their regulators’ third-party risk guidance. A commercial lender may not hold consumer data at all, but borrower and guarantor financials are still confidential under the loan documents and any NDA, so the same vendor diligence applies.
What should never go into an AI tool?
- Anything on a free or personal account.
- Social Security, social insurance and account numbers, unless the firm has approved that tool for that data.
- Documents your loan agreements or NDAs restrict, unless the vendor is covered by them.
What can AI do with loan documents?
Summarise rent rolls and operating statements, compare a draft term sheet against the credit approval, list conditions still outstanding, and flag missing documents. Each of those is a first pass that a credit analyst checks. Credit decisions, and anything sent to a borrower, stay with a person.
How Broker AI fits
Broker AI checks what the firm’s AI drafts for borrowers and partners against the firm’s own rules, records which AI tools are approved and who accepted the AI policy, and keeps the record of every check. It never stores the loan file; that stays in the firm’s own AI account. The live industry pack is mortgage; the same governance carries over to other lenders.
General information, not legal or compliance advice. Rules differ by state, province and regulator; check with your compliance team or counsel.
Questions people ask
- Does Claude train on documents a business uploads?
- Anthropic says inputs and outputs from its commercial products are not used for training by default, except feedback users choose to send, which owners can turn off. Consumer plans follow different terms.
- Is ChatGPT Enterprise safe for loan files?
- OpenAI says its business products don’t train on your data by default. “Safe” still depends on your contract, retention settings, who has access, and a person checking each output.
- Can AI make a credit decision?
- It can prepare the analysis, but decisions should stay with a credit officer, and consumer lending adds fair-lending and adverse-action duties an AI cannot carry.
- Is this legal advice?
- General information, not legal or compliance advice. Rules differ by state, province and regulator; check with your compliance team or counsel.
By DisruptionIO, the team behind Broker AI. Published October 28, 2026.