Onyrisanitize
Tools & AI6 min read

Does AI train on your data? Free, paid, and opt-out

Do your conversations train the AI? Often yes by default on free tiers. Why opt-out isn't enough, and the rule that actually protects you.

By Pierre de ONYRI

On many consumer assistants, your conversations can be used to improve the model — often on by default on free tiers. You can usually opt out, and paid or “enterprise” tiers generally don't train on your data. But opt-out only governs training: your prompt still leaves your machine and can be retained for other reasons. The robust rule: don't rely on a setting, and anonymize sensitive data before sending.

Training: often on by default

It's the point savvy users keep repeating: “it's on by default, you can turn off training, but you have to do it.” On free tiers, conversation content can be reused to train future models unless you change the setting. Many don't know that, or never do it.

Diagram: a prompt feeding a model with a training loop, and a partial, dashed opt-out toggle.
Opt-out cuts the training loop — but the prompt has already left your machine.

Opt-out only covers training

Turning off training helps, but it's partial protection. Once sent, the prompt can be logged, moderated or retained for security or legal reasons — regardless of the training setting. And as one clear-eyed comment notes: “we don't have any real visibility into what they do with it.” Opt-out asks you to trust a promise you can't verify.

  • Training: what opt-out cuts (in theory).
  • Logging and moderation: may remain for security and abuse.
  • Legal retention: some data is kept for a time by obligation.
  • Visibility: you can't audit what happens server-side.

Free, paid, API: what changes (and what doesn't)

Paid and “enterprise” tiers generally don't reuse your content for training — that's often the real value of the subscription, more than the power. But one thing never changes, whatever the plan: the data leaves your environment to be processed. Paying reduces some risks; it doesn't put your sensitive data back under your control.

The robust rule: don't depend on a setting

  1. 1Detection: an engine spots sensitive data in your prompt.
  2. 2Tokenization: each is replaced with a reversible token, kept in local memory.
  3. 3Sending: only the anonymized text goes out — training or not, it identifies no one.
  4. 4Restoration: the answer is de-tokenized in your browser.

ONYRI Sanitize applies this rule: detection and the token ↔ value mapping stay in your browser, and only the anonymized text reaches the model. Whether training is on or off, there's nothing identifying to learn.

Frequently asked questions

Does AI really train on my conversations?
On many consumer plans, yes by default, especially free ones: content can be used to improve future models. You can usually opt out in the settings, but it's a manual toggle — and many never do it.
Is turning off training enough to protect me?
No. Opt-out only covers training. The prompt still leaves your machine and can be logged, moderated or retained for other reasons. For sensitive data, the effective protection is not to send it — so anonymize it first.
Does the paid plan or API train on my data?
Generally no: paid and enterprise tiers don't reuse your content for training, and that's often what you're paying for. But the data still leaves your environment. Anonymization stays useful at any price.

Sources & references

Keep your sensitive data in your browser

ONYRI Sanitize detects and masks your sensitive data before it reaches the AI, then restores the answer — from names to API keys.

Anonymize my prompt

Read next