Your Vendor's New AI Feature May Use Customer Data by Default

The Feature Shipped With the Switch Already On

Some vendors add AI features to tools that already hold customer tickets, notes, and internal documents. Depending on the product and plan, data from those tools may be used to improve apps or AI features by default. That can include a setting an administrator needs to review. The defaults and the data covered vary, so check the actual product settings and terms before assuming what your team shared.

The conventional take is that this is a privacy story. It is not, or not only. It is a procurement story, and the number that matters is not a fine. It is the cost of the audit you now have to run to find out what your team already fed into a system you cannot claw it back out of.

Here Is What Is Actually Happening

Two real examples, both from 2026, both from vendors you have heard of.

Atlassian says it began using eligible Jira, Confluence, and Jira Service Management metadata and in-app data to improve its apps and AI experiences for all customers on August 17, 2026. Its settings differ by plan: Free and Standard always contribute metadata, though administrators can turn off in-app data; Enterprise administrators can turn off metadata too. Atlassian says it de-identifies and aggregates contributed data. This is a specific product policy, not proof that every customer note is fed verbatim into a shared model.

Salesforce's Customer Data Access setting controls whether eligible organizations share customer data for certain global predictive AI models and service improvement. Salesforce documentation says this access is allowed by default for eligible organizations and can be turned off by an administrator in Setup. The setting applies to specific services; it is not evidence that every Spring '26 generative AI feature trains on every CRM record.

The Distinction Your Team Missed

There are two different things a vendor can do with your data, and they get blurred on purpose. The first is running the model on your data to give you an answer, summarizing your ticket, drafting your reply. That is the feature working. Almost nobody objects to that.

The second is contributing data to a vendor's broader product or model-improvement work. That can include training, depending on the service and its terms; it is different from processing a ticket to answer your query. If information is included in model training, deleting the original record may not remove its influence from an already-trained model. The vendor's terms may describe both activities as "improving your experience," but the purposes and safeguards differ.

Run the Number Before the Renewal

The good news, if you want to call it that, is that the enterprise market has already priced this. A cautious buyer can ask for opt-in training, a written limit on secondary uses, and a clear retention period. Some business AI plans promise not to train on business data by default, but retention and license terms still vary by product. Read the terms for each tool instead of assuming that a business label settles the question. This shows that a vendor can offer different data terms at different tiers.

So the arithmetic for a small operator is this. Log into every SaaS tool that added an AI feature this year. Find the data or privacy settings, not the marketing page. Confirm whether training is on, and whether you are on a tier that can turn it off. If the opt-out lives behind an upgrade, that upgrade is now a real line item, and you get to decide whether the feature is worth what it costs to keep your data yours.

What the Sales Deck Leaves Out

The deck will tell you the AI feature saves your team four hours a week. It will not tell you the feature's training toggle defaulted to on, that the opt-out is tier-gated, or that "improve our services" is doing a lot of quiet work in the terms you clicked through. None of that is illegal. It is just the part of the deal that was structured to be discovered late, if at all. The vendor is betting you won't look. The fix costs one afternoon and a list of your tools. Go look.