AI inventory

Vertex AI

Google · Cloud AI platform · Confirmed · catalog vertex-ai. Seen by 1 lane through 1 finding.

Policy · none set
Policy changes re-score this app on the next sync, which starts right away.
Combined score
67
High one lane
People
7
Distinct identities across every lane
Seen by
1
Google Cloud
First seen
Aug 20, 2026
Last seen 18 h ago
gcpdenied-attemptsbuildersinvocationsservice-account-heavytuned-modelsnotebooksmany-projects
Evidence

Where it was seen

Strongest finding →
Google CloudGoogle Cloud (demo)
High67

55 audit entries by 7 principals, 5 denied, 4 resources in 6 projects

Audit entries
55
Principals
7
Projects
6
API-key calls
0
7 people · 18 h agoOpen finding →
Next steps

Recommended actions

  1. 01
    5 calls were denied by IAM. People are trying to reach Vertex AI; grant it through a reviewed role or add an organization policy restricting the service so the attempts stop.
    Google Cloud
  2. 02
    Tuned models or custom training jobs exist. Confirm what data trained them, that the source bucket is classified, and that model artifacts are not shared across projects.
    Google Cloud
  3. 03
    Notebook instances or runtimes exist. Check that they run without external IPs and inside a VPC Service Controls perimeter so datasets loaded into them cannot leave.
    Google Cloud
  4. 04
    Vertex AI is active in 6 projects. Consolidate AI work into a small set of governed projects with an owner, or use organization policies to restrict which projects may enable it.
    Google Cloud
Who

People

PersonDepartmentSourcesSeen bySign-ins · 30 d
chat-backend (service account)
n/a
Google Cloud principal
Google Cloud0
n/a
Google Cloud principal
Google Cloud0
n/a
Google Cloud principal
Google Cloud0
n/a
Google Cloud principal
Google Cloud0
n/a
Google Cloud principal
Google Cloud0
n/a
Google Cloud principal
Google Cloud0
scoring-job (service account)
n/a
Google Cloud principal
Google Cloud0
Vendor

Policy notes

Trains on data
No
Enterprise tier
Available
Org policy
None set

Google's managed ML and foundation-model platform inside the customer's project boundary; risk is who can call it, tuned models, notebooks, and whether Data Access logs are on.

Billing

What it costs

Last 30 days
$2,827
Previous 30
$2,411
Lanes

Score by source

Each lane scores its own evidence. The combined score takes the strongest lane and adds five points per additional lane, capped at 100.