0 / 0
Security and privacy for foundation models

Security and privacy for foundation models

Your work with foundation models is secure and private, in the same way that all your work on watsonx is secure and private.

Foundation models that you interact with through watsonx are hosted in IBM Cloud. Your data is not sent to any third-party or open source platforms.

The foundation model prompts that you create and engineer in the Prompt Lab or send by using the API are accessible only by you. Your prompts are used only by you and are submitted only to models you choose. Your prompt text is not accessible or used by IBM or any other person or organization.

You control whether prompts, model choices, and prompt engineering parameter settings are saved. When saved, your data is stored in a dedicated IBM Cloud Object Storage bucket that is associated with your project.

Data that is stored in your project storage bucket is encrypted at rest and in motion. You can delete your stored data at any time.

Privacy of text in Prompt Lab during a session

Text that you submit by clicking Generate from the prompt editor in Prompt Lab is reformatted as tokens, and then submitted to the foundation model you choose. The submitted message is encrypted in transit.

Your prompt text is not saved unless you choose to save your work.

Unsaved prompt text is kept in the web page until the page is refreshed, at which time the prompt text is deleted.

Privacy and security of saved work

How saved work is managed differs based on the asset type that you choose to save:

  • Prompt template asset: The current prompt text, model, prompt engineering parameters, and any prompt variables are saved as a prompt template asset and stored in the IBM Cloud Object Storage bucket that is associated with your project. Prompt template assets are retained until they are deleted or changed by you. When autosave is on, if you open a saved prompt and change the text, the text in the saved prompt template asset is replaced.

  • Prompt session asset: A prompt session asset includes the prompt input text, model, prompt engineering parameters, and model output. After you create the prompt session asset, prompt information for up to 500 submitted prompts is stored in the project storage bucket where it is retained for 30 days.

  • Notebook asset: Your prompt, model, prompt engineering parameters, and any prompt variables are formatted as Python code and stored as a notebook asset in the project storage bucket.

Only people with Admin or Editor role access to the project or the project storage bucket can view saved assets. You control who can access your project and its associated Cloud Object Storage bucket.

Logging and text in the Prompt Lab

Nothing that you add to the prompt editor or submit to a model from the Prompt Lab or by using the API is logged by IBM. Messages that are generated by foundation models and returned to the Prompt Lab also are not logged.

Ownership of your content and foundation model output

Content that you upload into watsonx is yours.

IBM does not use the content that you upload to watsonx or the output generated by a foundation model to further train or improve any IBM developed models.

IBM does not claim to have any ownership rights to any foundation model outputs. You remain solely responsible for your content and the output of any foundation model.

Learn more

Parent topic: Foundation models

Generative AI search and answer
These answers are generated by a large language model in watsonx.ai based on content from the product documentation. Learn more