AIDEFEND in Action AIA-0118 Incident

Vertex AI Resource Theft: About NT$30,000 in an Hour and the Limits of Budget Alerts

Physician and developer Lin reported about NT$30,000 in unauthorized Vertex AI charges within an hour across his Google Cloud projects. His first-person account, republished by Business Next, links those calls to unfamiliar service accounts (machine identities) created roughly two weeks earlier; Google subsequently adjusted most charges. The case concerns cloud AI resource theft, while the initial compromise path and any attacker use of AI agents remain unverified.

Resource HijackingIdentity & AccessCredential GovernanceVertex AICloud IAM
8 applicable AIDEFEND defenses

Threat Analysis

  • Paid model services were abused. Lin reported Gemini Flash and video-generation calls through Vertex AI, Google's managed AI platform. The traffic came from an IP address that resolved to Panama, which he took to be a relay. NT$30,000 describes abnormal billing; most charges were later adjusted.
  • Preparation preceded consumption. He described project enumeration, 17 AI APIs enabled in four seconds on his main project, and subsequent service-account and key creation. Other projects followed different sequences. The enumeration queries were spaced ten-odd minutes apart; the key was created and downloaded about two minutes after the service account. These are his readings of the logs, and the raw audit exports have not been independently examined.
  • Machine identities carried access. Service-account keys can obtain credentials within that identity's permissions. The exact IAM grants and token-issuance chain are unpublished; the described logs have not been independently reviewed.
  • Notification did not stop consumption. Google's early anomaly email arrived about an hour after the calls began; the US$50 monthly budget email he had configured arrived near noon, a delay he estimated at about five hours. Google documents that alerts-only budgets do not cap spending. The calls stopped only after he closed every billing account and shut down all of his projects.
  • Initial access is unconfirmed. He found Mac files dated about eight months earlier, including a LaunchDaemon entry that ran a hidden script which had since been deleted, created about two minutes after a visit to a site imitating the Claude Code install page. He infers that an infostealer took his browser session, which would explain the absence of a new-device alert despite two-step verification. Suspicious Mac files, a possible fake Claude Code coding-assistant site and stolen cookies remain a reconstruction. Huntress's separate Atomic macOS Stealer (AMOS) case cannot confirm this chain; attacker use of agents is unverified.

Applicable AIDEFEND Defenses (8)

AID-H-004.002
Service & API Authentication
High
Enforce short-lived workload authentication and least-privilege model access at Google Cloud, including direct Vertex AI calls. Organizations can additionally block new service-account keys and restrict covered access paths with VPC Service Controls. A separately protected administrator must govern these restrictions. Test approved-workload success and unauthorized-identity or outside-perimeter denial. The report does not establish the organization or perimeter prerequisites.
AID-H-004.001
User & Privileged Access Management
High
Remove unnecessary standing Owner/Editor and identity-administration rights from everyday accounts. Use separately protected, temporary, audited administration, with independent approval where supported. This limits an attacker holding only an everyday session from enabling services, creating identities or keys, and changing IAM. The report does not establish the exact roles held; these are preventive configurations.
AID-D-011.004
Non-Human Identity & Delegated Token Abuse Detection
High
Correlate identity creation, key issuance, IAM changes and API enablement with provider-side model invocations against an approved workload inventory. Include calls absent from application logs. Use Admin Activity logs for configuration changes and explicitly enable Data Access coverage for the actual invocation APIs. Test coverage and alert delivery, then route findings to separate containment. This can reveal unapproved access without guaranteeing advance detection.
AID-E-001.003
AI Agent & Workload Principal and Issuance Disablement
High
Preserve immutable principal IDs, IAM bindings and audit evidence, then disable confirmed unauthorized service accounts and issuance paths. Google's disable operation rejects existing access tokens and new token requests; verify both at the provider. Keep accounts disabled: re-enabling can restore acceptance of still-valid tokens. Secure the management path that could recreate access. This applies to workload identities without assuming an AI agent.
AID-E-001.001
Root & Long-Lived Credential Object Eviction
High
Preserve exact attacker-created key IDs and before/after inventories, disable those keys, then remove them after the evidence hold. Key deletion stops future use of that key but does not revoke previously issued tokens; keep the principal disabled and verify provider denial. Reset the administering human password from a clean device, and review issued sessions separately.
AID-I-003.003
Resource, Token & Spend Budget Throttling
Medium
Set provider spend caps per project and supported service, verifying the billed API is covered. Estimated-cost enforcement can lag and in-flight requests still accrue charges; caps reset monthly. Adjust only quotas actually exposed by the model/API, rather than assuming every Gemini quota can be set to zero. Remove unnecessary access through IAM. These provider-side constraints cover direct calls that application budgets cannot.
AID-D-005.007
Token, Tool-Use, Request-Parameter & Cost Spike Detection & Alerting
Medium
Detect project/caller usage and estimated-cost spikes against short-window baselines and budget thresholds; test notification delivery and correlate subsequent responses. Google Cloud Pub/Sub budget messages arrive several times daily and may be delayed, duplicated or reordered, so they are not a real-time spending gate. Automated billing shutdown is a separate response requiring service/data-impact review, not proof that detection stopped charges.
AID-E-001.002
Issued Token, Authentication Session & Lease Revocation
Medium
From a clean device, change the Google Account password, inspect devices and sessions, and terminate affected sessions. Google says password changes sign out most sessions, with exceptions; they help but do not establish complete eviction. Separately review app passwords and third-party authorizations, revoke unrecognized access, and verify session denial. Remediate the suspect Mac before reauthentication. Cookie theft remains unconfirmed.

What Defenders Should Do Now

  • From a clean device, change the administering Google Account password, review security events, devices, sessions, recovery settings and third-party authorizations, then terminate affected sessions and revoke unrecognized access. Keep the suspect Mac disconnected until remediation is complete.
  • Preserve project audit and billing evidence, including principal and key IDs, creation events, IAM changes, API enablement and model calls. Preserve the Mac files named in the source without executing them; their presence or absence alone does not establish the infection.
  • Check ownership before disabling unfamiliar service accounts. Preserve the evidence, disable confirmed unauthorized principals and their issuance paths, then disable and remove attacker-created keys. Verify that existing credentials and fresh token requests cannot reach the models, and secure the management path that could recreate access.
  • Remove unnecessary standing Owner/Editor and credential-management rights. Separate everyday work from protected, temporary administration; use short-lived workload credentials. Organizations can additionally evaluate policies blocking new service-account keys and covered access paths through VPC Service Controls.
  • Enable and test the model-call audit coverage required by the deployed APIs, correlate it with identity changes, and test alert delivery. Configure provider spend caps only where supported, verify scope and enforcement, and adjust only quotas actually available for each model. Automated billing shutdown needs a reviewed service/data impact and cannot guarantee an exact spending ceiling.

1 additional consideration

Forensic assessment and trusted recovery of the personal endpoint

Alongside the cloud-credential response, independently assess the personal computer used to access the accounts and recover it through a trusted endpoint process. One antivirus scan is not sufficient evidence that the computer is safe for renewed account access.
Recommendation: Preserve audit exports and suspicious files for qualified forensic analysis. Disconnect the suspect Mac and do not execute the files; complete endpoint cleanup or, where needed, reinstall from a trusted source before using it to sign in again.

Conclusion

This reported incident shows how unauthorized cloud access can become a model-service bill even when budget emails are configured. AIDEFEND  points teams toward provider-side workload authentication and access restrictions, separately protected administration, identity-aware usage detection, and distinct containment of service accounts, keys and sessions. Spend caps can help constrain supported services, but their reporting delay and coverage must be verified. The proposed AMOS-to-cloud infection chain and attacker use of agents remain unconfirmed.

Last updated: