Cloud Giants Reject Researcher's Confused Deputy Flaws Despite Privilege Escalation Risks
Microsoft and Google declined to officially recognize vulnerabilities that allowed attackers to become cluster-admin and Organization Owner while remaining invisible in audit logs.
Independent security researcher Justin O'Leary uncovered two confused deputy vulnerabilities in Microsoft Azure and Google Cloud Platform that could grant attackers full administrative control. Both tech giants declined to recognize the flaws as official vulnerabilities.
The Azure vulnerability resided in the Azure Kubernetes Service (AKS) backup service's Trusted Access feature. An attacker with the Backup Contributor role—an Azure RBAC position with zero Kubernetes permissions—could escalate to cluster-admin privileges. Microsoft rejected the vulnerability report, recommended MITRE not assign a CVE, and CERT/CC closed its tracking case citing CNA hierarchy rules. Microsoft silently patched the flaw without assigning a CVE, publishing a security advisory, or notifying customers, though the company officially denies making any product changes and claims the behavior was expected.
The GCP flaw involves Config Connector, which fails to perform authorization checks when a user submits an IAMPolicyMember referencing an external organization. An attacker with basic Kubernetes namespace access and no GCP permissions could become a GCP Organization Owner. Critically, these actions are logged as service account activity rather than the attacker's identity, making them essentially invisible in cloud audit logs.
"The worst part, the part that people should be worried about, is you actually don't see the attacker in the cloud audit logs. The attacker is essentially invisible," O'Leary said.
Google's Vulnerability Reward Program panel initially accepted the report as P1/S1 severity on March 27, 2026, with an engineer stating "Nice catch!" and indicating a bounty would be paid. On April 7, 2026, a Google Security Bot reversed the decision, stating the issue did not qualify for a reward because it was "working as intended." As of June 2026, the case remains marked as "in progress (accepted)" with no fix deployed and no CVE assigned.
Both vulnerabilities fall under CWE-441 (Unintended Proxy or Intermediary, also known as Confused Deputy) in MITRE's Common Weakness Enumeration. The confused deputy problem was coined by computer scientist Norm Hardy in a 1988 paper and describes scenarios where a high-privileged entity is tricked by a lower-privileged entity into performing an action the requester is not authorized for.
"My feeling is that this problem is still around, because when you're building these big cloud infrastructures, you're often doing it in a cookie-cutter approach," O'Leary said.
The vulnerabilities pose severe risks to enterprise and government resources, including FedRAMP-authorized environments, because attacks leverage legitimate service accounts to bypass firewalls and remain undetected in audit logs. Both companies declined to officially recognize the flaws as vulnerabilities requiring public disclosure.