AI Training Opt-Out

What is an AI training opt-out?

An AI training opt-out is a way to tell a vendor not to use your data to train its AI models. It can be a switch in your account settings, an email you send, or a line in your contract.

It only matters when the vendor's default is to train. If training is on unless you act, you need an opt-out. If training is off unless you act, the vendor needs your opt-in. For the full picture of how training clauses work, see our page on AI training on customer data.

Opt-in vs opt-out

• Opt-in: the vendor does not train on your data until you say yes. Silence protects you.
• Opt-out: the vendor trains on your data until you say no. Silence gives permission.

The difference is large in practice. With opt-out, every day before someone finds the setting is a day your data may be used.

Setting vs contract

An opt-out can live in two places, and they are not equally strong:

• An account setting is quick to use, but the vendor controls it. It can be renamed, moved, reset by a plan change, or removed in a product update.
• A contract clause in your order form, DPA or AI terms is harder to change. A signed order form can also win over website terms through an order of precedence clause. OpenAI's Services Agreement, for example, puts the order form first.

What it looks like in a vendor's terms

A typical clause reads like this:

We may use Inputs and Outputs to train and improve our models. Workspace administrators can turn this off in Settings. Opting out applies to data submitted after the change and does not affect de-identified or aggregated data.

• "We may use" sets training as the default, so this is an opt-out model.
• "Workspace administrators" means a single employee cannot opt out alone. Someone with admin rights must do it.
• "After the change" means data the vendor already used stays used.
• "Does not affect de-identified or aggregated data" is a carve-out. Stripped or combined versions of your data may still be used.

Real opt-outs vary a lot. Slack's privacy principles tell workspace owners to email Slack with the subject "Slack Global model opt-out request." The same page says opting out still lets your data improve features within your own workspace.

A real change we caught

Venpo flagged a change in Mixpanel's terms: Mixpanel added 150% overage fees and removed the AI data opt-in. Under the old wording, customers had to opt in before Mixpanel could use AI inputs and outputs to improve its AI features. The new Supplemental Terms give Mixpanel that right for de-identified data, with no separate opt-in, while still saying it will not train or fine-tune on that data except for the customer's benefit.

The same thing can happen in the other direction. Venpo also reported that Lovable now trains its AI on your prompts and code by default, which turned an opt-in world into an opt-out one.

Why it matters to you

Defaults decide what happens to most data. If no one in your company knows a setting exists, an opt-out model means training is happening.

An opt-out may not cover everything. Possible gaps include data sent before you opted out, de-identified or aggregated data, feedback you give on AI answers, and other products from the same vendor. Feedback is one example: OpenAI's Services Agreement, for example, lets OpenAI use any feedback you give "without restriction or compensation."

Your customers may expect it. If you process your customers' data in a vendor's tool, your contracts with them may limit how that data is used. A vendor training on it can put you in breach of those promises.

Settings can drift. Plan upgrades, workspace mergers and new product launches can create new places where training is on by default. That ties closely to AI features enabled by default.

What to check

• Find out whether training is opt-in or opt-out on your exact plan, and read the terms that apply to that plan.
• Locate the opt-out: setting, email, form or contract clause. Write down who in your company owns it.
• Check what the opt-out excludes, such as de-identified data, aggregated data, feedback or past data.
• Check whether it covers every product and workspace you use from that vendor, including add-ons and beta features.
• Ask for the opt-out to be written into your order form or DPA, so a settings change cannot undo it.
• Recheck the setting after plan changes and major product updates.

How Venpo helps

Venpo monitors your vendors' terms, AI terms and privacy policies and emails you when an opt-in becomes an opt-out, or when an opt-out gains new carve-outs. The Fine Print Scan answers, for each vendor, whether it trains AI on your data.