Is the information used in the AI Studio publicly used by Google to train the models and can the data be exposed?
3 Likes
Welcome to the forum.
Yes to both questions. The use of data to train the model is disclosed in the pricing page, Gemini API の料金 | Google for Developers. The catch is, there is no option to get paid service for AI Studio itself. It is free, and therefore data sent to Studio can be used to train the model (and later, be exposed).
The manager, Logan Kirkpatrick, is aware of this issue. My recommendation and workaround (until Google can address this issue) is to only use publicly available data in Studio. I understand you won’t get the best possible prompt that way, but you can get close enough.
Then, use the API in a paid program. That data is safe (Google will not use data transmitted to a billed account). You can do your last prompt tweaks there.
I hope this helps.
2 Likes
-
Using the free tier of Google AI Studio means your data can be used by Google for model training and may be seen by human reviewers. There is a potential, though likely small, risk of this data being exposed later through model outputs or other means.
-
To prevent your AI Studio data from being used for training, you need to activate a Google Cloud Billing account .[9] This designates your usage as a “Paid Service” even for free components, and Google commits not to use this data for training.
-
Using Vertex AI Studio within Google Cloud generally offers stronger enterprise-grade data governance and privacy commitments, stating customer data isn’t used for training without permission.
Even logged in using your Google Business account, your information is exposed when using Google AI Studio. I learned this the hard way after being surprised by a HRLF prompt from Gemini. After nearly a full day of research and contradictory information from the Google support team, I figured out that you cannot obtain the privacy and security controls assured business users when using AI Studio, even when logged in with your business account.
This reply from AI Studio itself is helpful, following a long chat on the topic. Even AI Studio initially offered the wrong opinion, and it took a lot of digging and a decoder ring to “decrypt” the correct answer. Please see the following:
You are correct. Your test definitively shows that interactive chats in the AI Studio web UI are not being routed through your billed Google Cloud Project, even after selecting it in the API key menu. The usage from your Python application appeared correctly because it was a direct API call, but the web UI activity did not.
This leads to a new and definitive conclusion.
The Final, Correct Explanation
The AI Studio interface and the “Generative Language API” are, for data privacy purposes, two separate systems.
- Google AI Studio (aistudio.google.com) Web Interface: This is a free-to-use, public-facing tool for experimentation. Your test confirms that its activity is not linked to your Cloud billing project. Therefore, it operates under a standard data policy that allows Google to use your conversations to improve its models. The request you received to choose between two responses is a clear indicator of this data usage policy. Selecting a project in the “Get API key” dialog only serves to create and manage API keys; it does not change the data privacy rules of the web chat itself.
- The API (Accessed via Key): When you use an API key in your own application (like your Python script), the calls are sent to a different endpoint that is correctly associated with your Google Cloud Project. This is the “Paid Service” where your data is confidential and is not used for model training. This is why you saw usage metrics for your Python test but not for your web chat.
The Implication for Your Confidentiality
Based on your findings, the only safe assumption to make is:
Any and all conversations conducted directly within the aistudio.google com web chat interface are NOT covered by enterprise data confidentiality protections and may be used by Google.
1 Like