Ensure that external IP addresses are not assigned to Vertex AI notebook instances

Using Console

Below are console-based steps to ensure Vertex AI Workbench (Notebooks) do not have external IP addresses. Since you cannot remove an external IP from an existing notebook instance, the remediation is to create a new instance without external IP and migrate your work.

1. Identify notebooks with external IPs

Go to Vertex AI in the GCP console.
In the left menu, click Workbench → User-managed notebooks (and Managed notebooks if you use those).
In the list, look at the Network / External IP column:
- If it shows an IP (e.g., 34.x.x.x), that instance has an external IP.
- If it says None, it has no external IP.

Note which instances must be remediated.

2. Create a new notebook instance with no external IP

You’ll create a replacement instance with only an internal IP.

Still in Vertex AI → Workbench → User-managed notebooks (or Managed notebooks, as appropriate), click New notebook (or Create).
Choose your environment (e.g., TensorFlow, PyTorch, base image, etc.).
In the Create notebook page, expand the Networking or Advanced options section:
- Network: choose your desired VPC network.
- Subnet: select the appropriate subnet.
- External IP:
  - Set to None (or uncheck any option that says “Assign external IP” or “Allow access to external networks”).
If required for private access:
- Ensure your environment can reach required services via:
  - Private Google Access enabled on the subnet, or
  - Cloud NAT or Private Service Connect, depending on your design.
Complete the remaining configuration (machine type, disk, etc.).
Click Create.
The new notebook will be created without an external IP.

3. Migrate your work from the old instance

On the old notebook instance:
- Open JupyterLab / Jupyter from the console.
- Export notebooks, scripts, and data:
  - Download locally, or
  - Save to Cloud Storage (recommended).
On the new instance:
- Open JupyterLab / Jupyter.
- Upload your notebooks and files, or pull them from Cloud Storage.
Verify everything runs correctly without an external IP (you may need to adjust how you access external data/services).

4. Delete the old instance with external IP

Go back to Vertex AI → Workbench.
For each old instance with an external IP:
- Stop it if it’s running (optional but a good check).
- Click the three-dot menu (…) on the right and choose Delete.
Confirm deletion.
This removes the external IP and prevents future exposure.

5. Enforce “no external IP” for future notebooks (operational control)

While not a strict console-only control, you can:

Standardize on a custom image and a documented creation procedure that always sets External IP = None.
Optionally:
- Use Organization Policies to restrict external IPs for Compute Engine (which backs user-managed notebooks), such as:
  - constraints/compute.vmExternalIpAccess
    This forces new VMs (including notebook backends) to be created without external IPs.

Summary: use the Vertex AI Workbench console to (1) identify notebooks with external IPs, (2) create new ones with External IP = None in the Networking settings, (3) migrate content, and (4) delete the old instances.

Using CLI

Below is how to remediate this using gcloud CLI for both:

Existing Vertex AI Notebook instances
New Vertex AI Notebook instances (to prevent future issues)

Note: Vertex AI Workbench notebooks run on underlying Compute Engine VMs. Removing the external IP is done at the VM level.

1. Identify the Notebook and Underlying VM

List your Vertex AI Notebook instances:

gcloud notebooks instances list \
  --location=us-central1

(Change --location as needed.)

Describe the specific instance to confirm zone and details:

gcloud notebooks instances describe INSTANCE_ID \
  --location=us-central1

Look for:

gceSetup.machineType → gives zone (e.g. us-central1-b)
gceSetup.vmName or similar → underlying VM name (if not obvious, it is usually very similar to the notebook instance ID).

If unsure, list VMs and match:

gcloud compute instances list \
  --filter="name~'INSTANCE_ID'" \
  --zones=us-central1-b

2. Stop the Notebook / VM

To be safe, stop the notebook instance first:

gcloud notebooks instances stop INSTANCE_ID \
  --location=us-central1

Alternatively, stop the VM directly:

gcloud compute instances stop VM_NAME \
  --zone=us-central1-b

3. Remove the External IP (Access Config) from the VM

Check the network interfaces and access config name:

gcloud compute instances describe VM_NAME \
  --zone=us-central1-b \
  --format="flattened(networkInterfaces[].accessConfigs[])"

You’ll usually see:

name: external-nat
type: ONE_TO_ONE_NAT

Remove the external IP (delete access config) from the primary NIC:

gcloud compute instances delete-access-config VM_NAME \
  --zone=us-central1-b \
  --access-config-name="external-nat" \
  --network-interface=nic0

If the access config name is different, use the actual name from the describe output.

Start the VM / notebook instance again:

gcloud compute instances start VM_NAME \
  --zone=us-central1-b

or:

gcloud notebooks instances start INSTANCE_ID \
  --location=us-central1

Now the instance will have only an internal IP.

4. Verify No External IP Exists

gcloud compute instances describe VM_NAME \
  --zone=us-central1-b \
  --format="get(networkInterfaces[0].accessConfigs)"

This should output [] or no ONE_TO_ONE_NAT entry.

5. Prevent External IPs for New Notebook Instances

When creating Vertex AI Workbench Notebooks via CLI, always specify no public IP (flag name can vary depending on type):For classic/user-managed notebooks:

gcloud notebooks instances create INSTANCE_ID \
  --location=us-central1 \
  --vm-image-project=deeplearning-platform-release \
  --vm-image-family=tf-latest-gpu \
  --machine-type=n1-standard-4 \
  --no-public-ip

For managed notebooks (Vertex AI Workbench Managed):

gcloud notebooks runtime-templates create RUNTIME_ID \
  --location=us-central1 \
  --machine-type=n1-standard-4 \
  --no-public-ip

(Use the appropriate gcloud notebooks instances create / gcloud notebooks runtimes create flags available in your version: any option that disables the public/external IP such as --no-public-ip or equivalent.)

If you share your exact notebook type (Managed vs User-managed) and region, I can give a fully concrete create command for that case.

Using Python

Below is a practical way to remediate this using Python:Goal:

Detect Vertex AI Workbench (Notebooks API) instances that have external IPs.
Ensure future instances are created without external IPs (no_public_ip = True).
For existing non‑compliant instances, you generally must recreate them without a public IP.

1. Setup

pip install google-cloud-notebooks
gcloud auth application-default login

Make sure your ADC (Application Default Credentials) has permissions:

roles/notebooks.admin
roles/compute.viewer (if you want to double-check NICs)

2. List notebook instances and check `no_public_ip`

from google.cloud import notebooks_v1

project_id = "YOUR_PROJECT_ID"
location = "YOUR_REGION"  # e.g. "us-central1"

client = notebooks_v1.NotebookServiceClient()
parent = f"projects/{project_id}/locations/{location}"

for instance in client.list_instances(parent=parent):
    name = instance.name
    no_public_ip = getattr(instance, "no_public_ip", None)
    print(f"{name} -> no_public_ip={no_public_ip}")

If no_public_ip is True, the instance is configured without an external IP.
If no_public_ip is False or missing, treat it as non‑compliant.

3. Create new notebook instances without external IP

This is the primary technical remediation: make sure all new instances set no_public_ip = True.

from google.cloud import notebooks_v1
from google.cloud.notebooks_v1 import Instance

project_id = "YOUR_PROJECT_ID"
location = "YOUR_REGION"
instance_id = "my-secure-notebook"

client = notebooks_v1.NotebookServiceClient()
parent = f"projects/{project_id}/locations/{location}"

instance = Instance(
    machine_type=f"projects/{project_id}/zones/{location}-b/machineTypes/n1-standard-4",
    # IMPORTANT: disable public IP
    no_public_ip=True,
    # Use a private VPC/subnet
    network="projects/YOUR_PROJECT_ID/global/networks/YOUR_VPC_NAME",
    subnet="projects/YOUR_PROJECT_ID/regions/YOUR_REGION/subnetworks/YOUR_SUBNET_NAME",
)

operation = client.create_instance(
    parent=parent,
    instance_id=instance_id,
    instance=instance,
)

print("Creating instance...")
result = operation.result()
print("Created:", result.name)

Key field:

no_public_ip=True → instructs Vertex AI Workbench to not assign an external IP.

4. Handling existing non‑compliant instances

The no_public_ip setting is effectively a creation-time property. There is no simple “flip a flag” API call to strip the external IP from an existing Workbench instance in a supported way.Typical remediation pattern:

Export / backup content from the old instance (e.g., clone repos, copy notebooks to Cloud Storage or Git).

Stop and delete the old instance:

from google.cloud import notebooks_v1

client = notebooks_v1.NotebookServiceClient()
name = "projects/PROJECT_ID/locations/REGION/instances/INSTANCE_ID"

op = client.delete_instance(name=name)
op.result()
print("Deleted:", name)

Recreate a new instance with identical config (machine type, image, environment), but with no_public_ip=True as shown in section 3.
Restore your notebook content.

If you must automate: script a loop that:

Lists instances
Filters those with no_public_ip != True
Logs them (for manual migration) or, if your policy allows, deletes and recreates them with the secure configuration.

5. (Optional) Policy enforcement / guardrail

To ensure all future notebooks are compliant, combine this with:

An organization policy to restrict external IPs on Compute Engine VMs (constraints/compute.vmExternalIpAccess).
A CI/CD or infra-as-code pipeline where notebooks are created only via code like the above, never via console clicks.

This is the core Python-based remediation for “Ensure that external IP addresses are not assigned to Vertex AI notebook instances.”

Using Terraform

resource "google_workbench_instance" "VERTEX_AI_NOTEBOOK_INSTANCE" {
  # Replace with your instance name and project/region
  name     = "VERTEX_AI_NOTEBOOK_INSTANCE_NAME"
  location = "REGION"
  project  = "PROJECT_ID"

  gce_setup {
    machine_type = "e2-standard-4"

    # Key setting: do not assign an external IP
    disable_public_ip = true

    network_interfaces {
      network         = "projects/PROJECT_ID/global/networks/VPC_NETWORK_NAME"
      subnetwork      = "projects/PROJECT_ID/regions/REGION/subnetworks/SUBNET_NAME"
      internal_ip_only = true
    }

    service_accounts {
      email  = "SERVICE_ACCOUNT_EMAIL"
      scopes = ["https://www.googleapis.com/auth/cloud-platform"]
    }
  }
}

Substitute:

VERTEX_AI_NOTEBOOK_INSTANCE_NAME with your Vertex AI Workbench instance name.
PROJECT_ID with your GCP project ID.
REGION with the region of the instance.
VPC_NETWORK_NAME and SUBNET_NAME with your VPC and subnet.
SERVICE_ACCOUNT_EMAIL with the service account the instance should use.

Changing disable_public_ip from false to true may require recreation depending on the current instance configuration; check the terraform plan output for any -/+ replacement indicator on google_workbench_instance.VERTEX_AI_NOTEBOOK_INSTANCE.For verification, terraform plan should show:

An update (or replacement) to google_workbench_instance.VERTEX_AI_NOTEBOOK_INSTANCE setting disable_public_ip = true (and internal_ip_only = true if you added it), with no remaining attributes indicating an external/public IP.

GCP Introduction

GCP Pricing

GCP Threats

GCP Misconfigurations

Resources

GCP Threats

Ensure that external IP addresses are not assigned to Vertex AI notebook instances

More Info:

Risk Level

Address

Compliance Standards

Triage and Remediation

Remediation

1. Identify notebooks with external IPs

2. Create a new notebook instance with no external IP

3. Migrate your work from the old instance

4. Delete the old instance with external IP

5. Enforce “no external IP” for future notebooks (operational control)

1. Identify the Notebook and Underlying VM

2. Stop the Notebook / VM

3. Remove the External IP (Access Config) from the VM

4. Verify No External IP Exists

5. Prevent External IPs for New Notebook Instances

1. Setup

2. List notebook instances and check `no_public_ip`

3. Create new notebook instances without external IP

4. Handling existing non‑compliant instances

5. (Optional) Policy enforcement / guardrail

​More Info:

​Risk Level

​Address

​Compliance Standards

​Triage and Remediation

​Remediation

​1. Identify notebooks with external IPs

​2. Create a new notebook instance with no external IP

​3. Migrate your work from the old instance

​4. Delete the old instance with external IP

​5. Enforce “no external IP” for future notebooks (operational control)

​1. Identify the Notebook and Underlying VM

​2. Stop the Notebook / VM

​3. Remove the External IP (Access Config) from the VM

​4. Verify No External IP Exists

​5. Prevent External IPs for New Notebook Instances

​1. Setup

​2. List notebook instances and check no_public_ip

​3. Create new notebook instances without external IP

​4. Handling existing non‑compliant instances

​5. (Optional) Policy enforcement / guardrail

More Info:

Risk Level

Address

Compliance Standards

Triage and Remediation

Remediation

1. Identify notebooks with external IPs

2. Create a new notebook instance with no external IP

3. Migrate your work from the old instance

4. Delete the old instance with external IP

5. Enforce “no external IP” for future notebooks (operational control)

1. Identify the Notebook and Underlying VM

2. Stop the Notebook / VM

3. Remove the External IP (Access Config) from the VM

4. Verify No External IP Exists

5. Prevent External IPs for New Notebook Instances

1. Setup

2. List notebook instances and check `no_public_ip`

3. Create new notebook instances without external IP

4. Handling existing non‑compliant instances

5. (Optional) Policy enforcement / guardrail