A data scientist is using Vertex AI Workbench and needs to share a notebook with a team member who only needs read access. What is the most secure way to do this?
Trap 1: Add the user as a 'Vertex AI User' on your project.
Granting project-level Vertex AI User lets the member launch and access all notebooks and resources, exceeding read-only sharing of one notebook. It is tempting because that role is the standard way to give a colleague working access to Vertex AI, which fits collaboration requiring execution rather than passive viewing.
Trap 2: Copy the notebook code into an email.
Emailing code strips the notebook's execution environment, outputs, and metadata, and exposes sensitive data outside Google Cloud's access controls. It is tempting because it is quick and needs no permissions, which suits sharing a small, non-sensitive snippet with someone lacking any project access.
Trap 3: Change the instance permissions to 'Public'.
Public permissions expose the notebook to anyone on the internet, granting far more than the single team member's read access. It is tempting because it removes all IAM configuration, which suits deliberately publishing a sanitised notebook intended for unrestricted public consumption.
- A
Add the user as a 'Vertex AI User' on your project.
Why it fails: Granting project-level Vertex AI User lets the member launch and access all notebooks and resources, exceeding read-only sharing of one notebook. It is tempting because that role is the standard way to give a colleague working access to Vertex AI, which fits collaboration requiring execution rather than passive viewing.
- B
Grant the user 'Storage Object Viewer' role on the Cloud Storage bucket containing the .ipynb file.
Granting Storage Object Viewer on the bucket gives read-only access to the .ipynb object without granting broader project or notebook editing rights. This satisfies the stem's least-privilege requirement for a team member needing only read access.
- C
Copy the notebook code into an email.
Why it fails: Emailing code strips the notebook's execution environment, outputs, and metadata, and exposes sensitive data outside Google Cloud's access controls. It is tempting because it is quick and needs no permissions, which suits sharing a small, non-sensitive snippet with someone lacking any project access.
- D
Change the instance permissions to 'Public'.
Why it fails: Public permissions expose the notebook to anyone on the internet, granting far more than the single team member's read access. It is tempting because it removes all IAM configuration, which suits deliberately publishing a sanitised notebook intended for unrestricted public consumption.