GCP Generative AI Leader Practice Question 44: Customer Data Not Used for Training
Google Cloud Generative AI Leader · Test 5 · question 4 of 10
A hospital's general counsel is evaluating Gemini for Google Cloud for clinical staff to draft patient correspondence. Before approving the pilot, she asks whether the confidential prompts staff type and the draft text the model returns could ever be used to improve Google's general-purpose foundation models for other customers.
What should she be told about Google's default handling of this data?
Show answer and explanation
Answer:
Google's enterprise gen AI offerings are built on a commitment that customer data, including prompts and generated responses, is not used to train Google's foundation models. This is a default behavior, not something the hospital needs to opt into or a setting it needs to turn off, and it is one of the core reasons organizations in regulated industries can consider adopting these tools.
Option A is wrong because no retraining on the hospital's data happens at all, anonymized or not. Option B is wrong because there is no opt-in "shared improvement" program that would require the hospital's confidential prompts to train the general model. Option D is wrong for the same reason in reverse: there is no setting the hospital must remember to disable, because training on customer data is not the default in the first place.
Reference: How Gemini products in Google Cloud use your data