CVE-2025-30167: jupyter_core: config hijack enables cross-user code exec

## Impact On Windows, the shared `%PROGRAMDATA%` directory is searched for configuration files (`SYSTEM_CONFIG_PATH` and `SYSTEM_JUPYTER_PATH`), which may allow users to create configuration files affecting other users. Only shared Windows systems with multiple users and unprotected `%PROGRAMDATA%` are affected. ## Mitigations - upgrade to `jupyter_core>=5.8.1` (5.8.0 is patched but breaks `jupyter-server`) , or - as administrator, modify the permissions on the `%PROGRAMDATA%` directory so it is not writable by unauthorized users, or - as administrator, create the `%PROGRAMDATA%\jupyter` directory with appropriately restrictive permissions, or - as user or administrator, set the `%PROGRAMDATA%` environment variable to a directory with appropriately restrictive permissions (e.g. controlled by administrators _or_ the current user) ## Credit Reported via Trend Micro Zero Day Initiative as ZDI-CAN-25932

An attacker with a low-privileged Windows domain account on a shared ML research server writes a malicious jupyter_notebook_config.py to %PROGRAMDATA%\jupyter\. The file registers a kernel startup hook that silently exfiltrates environment variables (including OPENAI_API_KEY, AWS credentials, HuggingFace tokens) and scans for .ipynb files containing embedded secrets. A senior ML engineer logs in, launches their training notebook, and their credentials plus in-progress model checkpoints are exfiltrated to an attacker-controlled endpoint. On JupyterHub Windows deployments, a single compromised student or contractor account can persist access affecting all platform users until the malicious config is discovered.