Is it safe to put confidential CRE deals into AI?

Yes, when the setup is right. It is safe on an enterprise or API tier where the provider does not train on or retain your inputs, and on a self-hosted model where nothing leaves your environment. It is not safe in a free consumer chatbot on default retention. The tier you choose is the control, not the brand.

There are three tiers, and the difference is who can see your data. A free consumer chatbot processes your input under consumer terms and may retain it for product improvement unless you opt out. An enterprise or business API tier runs under a data-processing agreement where your inputs are not used to train the vendor's models by default. A self-hosted open-source model runs on infrastructure you control, so a rent roll, a T12, or an LP list never leaves your walls at all. Most CRE firms do not need to self-host; they need to stop pasting deals into the free app and move onto a no-train tier.

One honest caveat on the provider claims: data terms are set by each vendor and they change. Anthropic's Commercial Terms of Service state that Anthropic will not use your Inputs or Outputs to train its models (Anthropic Commercial Terms of Service, current as of 2026). OpenAI states that data submitted through its API is not used to train its models by default and business/enterprise data is excluded from training (OpenAI API data usage and Enterprise privacy policies, current as of 2026). Read the version dated to the day you sign, and treat this as architecture guidance, not legal advice; run your data policy past counsel before it becomes firm standard.

This is exactly what a NextAutomation build is designed around. We architect so your sensitive documents stay in your environment, connections are encrypted, and the model is chosen to meet your privacy requirement rather than whatever a broker happened to paste into. If self-hosting is warranted, we deliver on your own infrastructure; if a no-train API tier is sufficient, we configure and govern it for you. The Data, Security and Compliance FAQ covers the account-level specifics, the AI Underwriting Copilot shows what a grounded, in-your-environment system looks like on a live deal, and OpenAI vs Anthropic for enterprise compares the two no-train tiers head to head.

A pure brokerage that only ever handles already-public listing data has less to protect and can move faster with less scaffolding. Every other CRE firm, anyone touching LP data, underwriting assumptions, or off-market pipeline, should start with a data-protection review before a single deal goes near a model. A NextAutomation Operations Audit produces exactly that review and the safe architecture to go with it. That is the first step, and it is the recommended one.