Does Stability AI train on your data?
Stability AI trains on your Inputs and Outputs unless you opt out. The switch is "Training: Improve the Model for Everyone" in account settings; which way it ships is stated nowhere.
A self-serve opt-out exists
"To improve and develop our Services and conduct research, including to train and improve our models and develop new product features. When we use your Inputs and Outputs to train our models, you can opt-out of our use of this personal data for these training purposes."
- 10 Sept 2026
- 5 Oct 2026
- All accounts
- global
- Fetched from the company
- 1F9E58C6…2E18
How we read it: Re-graded from UNCLEAR on the privacy policy effective 31 July 2025, which replaced a 2024 document that never mentioned training. Inputs are "text, images, and audio, depending on the products you use". The policy's "Learn more" leads to the Privacy Center, which names the control per product: the "Training: Improve the Model for Everyone" toggle in web account settings and on the Platform API account page, "Disable History and Training" in Stable Assistant, and an account-page opt-out for Stable Audio, where uploaded audio is never trained on but generated audio may be. D rather than C because the switch's default position is stated nowhere. Nothing on human review or de-identification.
opt out, opt-out, object, settings, toggle, default, train
What the document does offer
- in the privacy policy, the sentence after the training purpose: "When we use your Inputs and Outputs to train our models, you can opt-out of our use of this personal data for these training purposes. Learn more"
- in the Privacy Center (stability.ai/privacy-center, captured 2026-09-10): "Can I opt out of the use of my inputs and outputs for training Stability AI's models?" followed by steps for the web account, the Platform API, Stable Assistant and Stable Audio
- in the Privacy Center: "Your uploaded audio is only used during your interactions with Stable Audio without becoming part of the model's training data. However, note that the generated audio resulting from your interactions (not your uploaded audio) may be used to improve and train the model."
- "default": 0 matches in either document
Why none of it applies: The control exists and is located: the policy says you can opt out and links the Privacy Center, which walks through the toggle for the web account, the Platform API and Stable Assistant, and a separate opt-out for Stable Audio outputs. What neither document says is the position a new account starts in. "Default" appears in neither; the instructions are written as turning something off. By this site's rule that is an undisclosed default, and the tier scores as a toggle whose position the policy does not state.
What each plan can refuse
The same questions, asked of every plan we could identify. Every cell comes from a quoted clause or a documented search - see the evidence above.
- Self-serve toggle
- Not stated
- 53 / 100
- Not stated
- Unknown
How to opt out of Stability AI training
Plan by plan, from what Stability AI's own policy says about refusing. Where it publishes the steps, they're listed here with the limits they carry. Where there's nothing to switch off, or nothing graded, this says so instead of inventing a menu path.
All accounts
Yes. Stability AI has a self-serve setting.
- Log in and select your avatar in the top right corner.
- In the drop-down, select the account Settings.
- Under User Preferences, set the Training, Improve the Model for Everyone toggle to Off.
- Go to platform.stability.ai/account/overview.
- Set the Training, Improve the Model for Everyone toggle to No.
- Open Profile, then Settings.
- Use the Disable History and Training toggle.
- Log in to StableAudio.com and open your account page from your avatar in the top right.
- Uploaded audio is not trained on at all; the opt-out covers the audio the service generates for you.
What this policy has done since we started watching
The record starts here. Stability AI's 2 documents were first captured on 10 Sept 2026, and there's no earlier copy to compare them against, so this verdict has no history behind it yet. From now on every re-read is diffed against those copies, and anything that moves appears here.
Vendor response
We wrote to Stability AI on 15 Sept 2026, quoting the clauses above and inviting a correction. Nothing has come back. That's published because a company declining to comment on its own policy is something a reader should know, and because the invitation should be on the record whether or not it's taken up. Any reply goes here, word for word.
Report a wrong clause