What happened
Replaying the Google Cloud billing takeover
- 1
Phishing for a privileged user
Date not statedA Google Search ad leads a privileged user to a fake Google sign-in page; the kit captures the password and the MFA code and signs in with them, then the actor signs in again from a second address.
- 2
Dormant projects and borrowed identities
Date not statedIn existing low-usage projects the actor enables the Gemini API, creates service accounts and Gemini API keys and gives the service accounts high-privilege roles, then deletes an organization policy that restricts grants to corporate domains.
- 3
Billing account takeover
Date not statedExternal Gmail accounts become billing administrators, the real administrators are demoted to viewers, and the actor's own projects are linked to the billing account.
- 4
Gemini usage billed to the victim
Within 24 to 48 hours of initial accessAutomated, high-volume Gemini API usage from the linked projects runs on the victim's bill; the victim cannot see the prompts, the outputs or the workloads.
In multiple intrusions into Google Cloud environments the actor took over billing accounts and linked its own projects, so the victims paid for the actor's Gemini API workloads: charges of tens to hundreds of thousands of dollars within 24 to 48 hours of initial access, per Wiz as of October 20, 2026. Victims could not see the prompts or the outputs. Google flagged the first sign-in as suspicious and emailed the user, and the session went on; the sources report no action of the actor as refused or blocked. The number of intrusions and projects, the campaign dates and the cost per victim are not stated.
Where cloud-provider controls fit
On Google Cloud, two controls would have refused a step. A Cloud Identity or Google Workspace 2-Step Verification policy that allows only security keys or passkeys for the privileged users refuses the relayed sign-in, so a password and a code typed into the fake page do not complete a sign-in. An IAM deny policy refuses the delete of an organization policy to every principal outside its exceptions, if the phished user is not an exception principal and cannot change deny policies. The other controls would only have limited the incident. Context-Aware Access keeps the Google Cloud console and APIs to a managed device or a trusted network. A service allowlist refuses the Gemini API in your own projects, and the allowed-members constraint refuses grants to outside principals while it stands, including on a billing account that belongs to the governed organization. Paying for projects in that organization alone does not establish that ancestry. A principal with the right permission can remove each of these three, and the actor deleted organization policies in this incident. A spend cap budget, generally available since October 13, 2026, pauses new Gemini API requests in one project for one service once its cost passes the amount. It is not instant, and a Billing Account Administrator, the role the actor held, can lift it. Two steps stay outside every control: the ad lure, and the link of outside projects to your billing account. No Google Cloud setting caps what a billing account pays across all its linked projects.
Could this have been prevented?
In hindsight, not as a whole. 2 controls would have refused a step outright, and 5 controls would have limited one. 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.
- 2
- Prevented
- 5
- 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 the Google Cloud billing takeover
1 · Phishing for a privileged user
1 Prevented1 Contained1 Not covered3 on other clouds2 · Dormant projects and borrowed identities
1 Prevented2 Contained6 on other clouds3 · Billing account takeover
1 Contained1 Not covered4 · Gemini usage billed to the victim
1 Contained1 Not coveredKey facts
- The actor bought Google Search ads for terms such as google cloud and gcp; a click led through a Google Sites redirect to a fake Google sign-in page, and Google Ads removed the campaign following reports, on a date Wiz does not state
- The kit captured the credentials and the MFA code, returned the user to the real Google prompt, which forced a second login, and signed in from 212.73.134.136 within moments; minutes later the actor signed in from a second address that Wiz does not name. Google flagged the first sign-in as suspicious and emailed the user; the session went on
- In existing, low-usage projects the actor enabled the Gemini API, created service accounts and Gemini API keys and gave the service accounts roles including roles/ml.admin; the number of projects is not stated
- The actor deleted organization policies, with iam.allowedPolicyMemberDomains given as an example, which allowed IAM grants to non-corporate accounts; which other policies were deleted is not stated
- External Gmail accounts under the actor's control were added to the billing account with admin roles, pre-existing billing administrators were demoted to billing viewers, and attacker-controlled projects were linked to the billing account
- Charges reached tens to hundreds of thousands of dollars for each victim within 24 to 48 hours of initial access, as of the Wiz report of October 20, 2026, and the victims could not see the prompts or outputs; Wiz states no victim names, no count of intrusions and no campaign dates
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.From Phishing to LLMjacking: Threat Actors Targeting Cloud Billing Accounts for LLM UsageWiz Customer Incident Response Team · Oct 20, 2026 · ResearchAttack chain, roles and permissions, audit events, cost range, indicators, unknowns
- 2.Cloud Billing Hijacking as LLMjacking: Security Implications and GuidanceCloud Security Alliance AI Safety Initiative · Oct 21, 2026 · ResearchAttack chain as restated, Google Cloud controls cited, billing and Gemini cap behavior, unconfirmed items
- 3.Reading Wiz CIRT's LLMjacking Report: The Flow of How a Google Cloud Billing Account Gets Hijacked and You End Up Paying for Someone Else's Gemini API Usage, and What to CheckClassmethod (DevelopersIO) · Oct 24, 2026 · ResearchAttack chain as restated, budget and alert behavior, billing role behavior
The estate in the replay is illustrative and drawn at the level of attack stages; the victim's real environment is not public. Verdicts describe what a control does when it is in place before the incident. They are not an assessment of the victim's security program.