Internet Access Whitelist Limits for Kaggle Notebooks

Users who rely on Kaggle Notebooks for reproducible research often need to call external APIs—for example, to fetch pretrained models, query public datasets, or interact with cloud services—while also wanting the session to survive the platform’s runtime limits. The notebook environment runs on Ubuntu 18.04 LTS and enforces a 9‑hour GPU or 12‑hour CPU maximum, after which the container is terminated and the /kaggle/working directory is cleared. However, outbound internet traffic is currently restricted to a whitelist of approved services, and the criteria for adding new endpoints remain under discussion, creating uncertainty about which external APIs can be relied upon without prior approval.

What factors does Kaggle evaluate when deciding to whitelist a new domain or API? Is there a public roadmap or timeline for expanding the whitelist to cover commonly used services? Can notebook authors request temporary access to specific external endpoints for a given session, and if so, what is the process?