Google Vertex AI Agents Vulnerable to Exploitation via Default Permissions
Security researchers identified a vulnerability in Google Cloud's Vertex AI Agent Engine where default permission configurations allowed attackers to compromise deployed AI agents, potentially exfiltrating customer data and accessing proprietary code. Google issued guidance recommending customers use their own service accounts instead of defaults to mitigate the risk.
What happened
Security researchers discovered that Google Cloud Platform's Vertex AI Agent Engine contained a vulnerability allowing attackers to exploit default permission scoping to weaponize deployed AI agents.
Attackers could turn AI agents into 'double agents' that exfiltrate data and compromise cloud infrastructure.
Reported impact
Customer data and proprietary Google code were exposed through the vulnerability.
Excessive default permissions allowed attackers to pivot and access Cloud Storage.
Index analysis
The vulnerability stems from over-privileged default service account configurations in Vertex AI agents.
Cloud misconfigurations in AI agent deployment created an attack surface for threat actors.
Governance controls
Google updated documentation to urge customers to use Bring Your Own Service Account (BYOSA) instead of default service accounts.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.