Critical Gemini CLI Flaw Enabled Host Code Execution Before Sandboxing

Summary
Researchers said Gemini CLI trusted workspace configuration that could run attacker-controlled commands on the host before sandboxing. Google patched Gemini CLI and the run-gemini-cli GitHub Action; the flaw could expose accessible secrets and enable CI/CD supply-chain.
Key points
- Novee Security researchers found that Gemini CLI loaded agent configuration from the current workspace without review, sandboxing, or human approval.
- Malicious configuration planted in the workspace could cause arbitrary host commands to run before sandbox initialization.
- Researchers said code execution could expose secrets, credentials, and source code available to the agent.
- They said an attacker could use the flaw to steal tokens, move laterally to downstream systems, or target CI/CD supply chains.
- The attack path did not involve prompt injection or a model decision.
- Google patched the vulnerability in Gemini CLI and the run-gemini-cli GitHub Action.
Article Details
- Vulnerability Types
- Remote code execution
- Severity
- Critical
- Exploitation Status
- unknown
- Exploit Availability
- unknown
- Patch Status
- available
MITRE ATT&CK
Vendors
Products
Gemini CLIA critical remote code execution vulnerability was recently discovered by researchers in Gemini CLI, an open source AI agent designed to provide lightweight access to Gemini directly from a terminal.run-gemini-cliThe vulnerability, patched by Google in both Gemini CLI and the ‘run-gemini-cli’ GitHub Action, was identified by researchers at Novee Security.
