Get hands-on experience with 20+ free Google Cloud products and $300 in free credit for new customers.

Privacy of Data on LLM

 Hi folks

Just wanted to know if there's any privacy doc available from Google stating that with the training of LLM models that we do on top of Google cloud LLMs does that data is taken by google for their model enhancement?

Thanks in advance

Solved Solved
2 9 13.3K
1 ACCEPTED SOLUTION

The document you linked is covering a service outside of Vertex AI. Essentially, there are two separate approaches to using a Google AI model: either outside of or inside of Vertex AI. Let's use the Palm API as an example.

  1. Outside of VertexAI, when you use PalmAPI, your usage is covered by these terms of service and the documents you linked.
  2. Inside of VertexAI, which is designed for enterprise-grade support, security and compliance, and service level agreement (SLA), you would be accessing the Vertex AI Palm API. Your usage is covered by Google Cloud's terms of service. There is more detail there about the training restriction: "Google will not use Customer Data to train or fine-tune any AI/ML models without Customer's prior permission or instruction" written in the additional terms of service.

So while the underlying model is Palm API for both methods, the terms of usage are different. Hopefully that helps clarify. 

 

 

View solution in original post

9 REPLIES 9

AndrewB
Community Manager
Community Manager

The policy outlining that Google Cloud does not use customer data to train its foundational models is documented hereMore details about how Google processes data can also be found in Google's Customer Data Processing Addendum (CDPA).

Thanks @AndrewB  Appreciate your response. but just wondering with the context in this doc here https://developers.generativeai.google/terms#content_license_and_data_use  when we use vertex AI/ any service for that matter, we will enable the APIs in the first place does that mean the license we are granting to Google under the “Submission of Content” ?

The document you linked is covering a service outside of Vertex AI. Essentially, there are two separate approaches to using a Google AI model: either outside of or inside of Vertex AI. Let's use the Palm API as an example.

  1. Outside of VertexAI, when you use PalmAPI, your usage is covered by these terms of service and the documents you linked.
  2. Inside of VertexAI, which is designed for enterprise-grade support, security and compliance, and service level agreement (SLA), you would be accessing the Vertex AI Palm API. Your usage is covered by Google Cloud's terms of service. There is more detail there about the training restriction: "Google will not use Customer Data to train or fine-tune any AI/ML models without Customer's prior permission or instruction" written in the additional terms of service.

So while the underlying model is Palm API for both methods, the terms of usage are different. Hopefully that helps clarify. 

 

 

Thank you for more deets and clarification @AndrewB  that's helpful 🙂

Hi @AndrewB , thanks for the above answer.
Can you explain how one accesses PalmAPI from both inside & outside of VertexAI.

Is there any more information availale about data privacy when using generate AI models ?

Be sure to register for the upcoming community webinar on AI Security & Privacy and you can submit your questions to the team in advance. 

I went to the terms of service

However I could not locate,  "Google will not use Customer Data to train or fine-tune any AI/ML models without Customer's prior permission or instruction" . Could you share the exact link, this is for a customer, which I need to share.

I think I found it, here : https://cloud.google.com/terms/service-terms

AviAIML_0-1704735014742.png

 

Does Google take consent from Google cloud users before enabling the vertex AI api ? I am asking because it is written that Google will not use Customer Data to train or fine-tune any AI/ML models without Customer's prior permission or instruction.