What the demonstration did
Replaying Tenable's ImageRunner demonstration / reported October 2024; the test date is not stated
- 1
Deploying a private image without reading it
Test date not stated · before October 19, 2024The deployer updates a running Cloud Run service with a revision that names a private image, and Cloud Run's service agent pulls it without checking the deployer's access.
- 2
A shell inside the private image
Test date not stated · before October 19, 2024The revision's container arguments open a reverse shell from the private image to the listener, through which the image's contents and secrets are open to inspection.
- 3
Report and fix
October 19, 2024 to January 28, 2025Tenable reports the flaw, and Google adds a deploy-time check that the deployer can read the image, fully rolled out on January 28, 2025.
In Tenable's own project, a new Cloud Run revision ran a private image of the same project and opened a reverse shell to the researcher's listener; reading the image's secrets is stated as possible, not observed, and no source reports exploitation. Google's fix was fully rolled out on January 28, 2025.
What ImageRunner shows about your Cloud Run deployers
Two permissions to update a Cloud Run service were enough to run a private image of the same project and open a reverse shell from it, because Cloud Run's service agent pulled the image without checking that the deployer could read it. Google closed that gap for every project: since January 28, 2025, the principal that deploys needs explicit access to the image. On your Google Cloud projects, an IAM deny policy on Cloud Run service updates bounds who can deploy, and Binary Authorization admits only images your pipeline attested; neither is shown to refuse Tenable's identity, and breakglass on a service update bypasses Binary Authorization. A reverse shell is refused only when all of a revision's traffic goes through your VPC network: an organization policy that requires all-traffic VPC egress keeps it there, and a hierarchical firewall policy on that network refuses the listener, although Google's pages do not name Cloud Run traffic for those policies. No control refuses a process reading its own image, so keep secrets out of the images you build.
Would the controls have stopped it?
In hindsight, no control would have stopped it. 4 controls would have limited how far it went. 3 steps have no cloud-provider control. The counts below are control assessments, not attack stages: several can address the same step. The same replay with the controls in place before the incident shows where each phase would have been refused or bounded.
- 0
- Prevented
- 4
- Contained
- 3
- Not covered
How it could have been prevented
The same replay with the controls in place. Green marks where a step is refused, blue where it is bounded, and red where no control reaches it. Open a row to read the control, why it acts, and its caveat.
Replaying Tenable's ImageRunner demonstration / reported October 2024; the test date is not stated
1 · Deploying a private image without reading it
2 Contained1 Not covered2 · A shell inside the private image
2 Contained1 Not covered6 on other clouds3 · Report and fix
1 Not coveredKey facts
- The deployer identity held run.services.update and iam.serviceAccounts.actAs and neither Artifact Registry Reader nor Storage Object Viewer, per Tenable
- Cloud Run's service agent pulled the private image with its own registry permissions and did not check that the deployer could read it; Tenable's advisory TRA-2025-04 put private images in Artifact Registry and Container Registry of the same project in scope
- The revision's container arguments ran netcat with /bin/bash and opened a reverse shell to the researcher's listener; Tenable said an attacker could then read the image's contents and secrets, which no source reports as observed
- Tenable reported the flaw on October 19, 2024; Google said it was working as intended on October 24, 2024, then sent a Mandatory Service Announcement to the owners of affected projects, folders and organizations in the last week of November 2024
- A Cloud Run release note of January 13, 2025, labeled Breaking, required the principal that creates or updates a resource to hold explicit permission to access the image; the change was 100% rolled out to production on January 28, 2025
- A Google Cloud spokesperson told SecurityWeek that an image from another project already needed an explicit IAM permission before the fix; Tenable rated the flaw Medium on February 18, 2025, and GovInfoSecurity reported no known exploitation on April 1, 2025
Sources
Every fact in this breakdown comes from the public reporting below. The replay is a model of that reporting, not a reconstruction of the environment it describes.
- 1.ImageRunner: A Privilege Escalation Vulnerability Impacting GCP Cloud RunTenable Research · Apr 1, 2025 · ResearchThe deployer's two permissions, the revision update, the image pull by the service agent, the reverse shell, what the shell could read, the November 2024 announcement and the January 28, 2025 rollout
- 2.Google Cloud Platform (GCP) Privilege Escalation Vulnerability in GCP Cloud Run (TRA-2025-04)Tenable Research · Feb 18, 2025 · ResearchThe disclosure timeline from October 19, 2024, the registries in scope, the Medium risk rating and the fix
- 3.Cloud Run release notes (entry of January 13, 2025)Google Cloud · Jan 13, 2025 · Vendor reportThe breaking change that requires the deployer to hold explicit permission to access the image
- 4.ImageRunner Flaw Exposed Sensitive Information in Google CloudSecurityWeek · Apr 2, 2025 · NewsThe Google Cloud spokesperson's statement on the fix and on the cross-project check that existed before it, and the notification and rollout dates
- 5.Google Cloud Fix Blocks Unauthorized Container AccessInformation Security Media Group (GovInfoSecurity) · Apr 1, 2025 · NewsNo known exploitation of the flaw
The estate in the replay follows the researchers' description of their own test environment. Verdicts describe what a control does when it is in place before the run. They are not an assessment of any company's security program.