Grok Build users agentic AI attack — Jul 2026
The Grok Build coding agent faced scrutiny after autonomously uploading sensitive user files including SSH keys and password databases to xAI servers
- Reported date
- Jul 16, 2026
- Target
- Grok Build users
- Agent type
- Coding agent
- Agent role
- Compromised agent
The exact incident date was not established. This entry is dated by its source report.
The xAI coding agent known as Grok Build faced significant criticism after users discovered the tool was autonomously uploading entire directory contents to xAI's Google Cloud servers without explicit authorization. The incident raised concerns regarding the handling of sensitive local data by automated development tools.
What happened
Grok Build is a terminal-based coding agent invoked via the grok command, designed to read and edit codebases, execute shell commands, search the web, and manage long-running tasks. It operates interactively, headlessly for scripting and CI, or embedded in editors via the Agent Client Protocol. Users reported that the agent transferred sensitive files, including SSH keys, password databases, documents, and photos, to xAI's infrastructure.
In response to the discovery, xAI disabled the upload feature and Elon Musk announced that all previously uploaded user data would be fully deleted. According to the company, data storage had been set to off by default starting July 12. To address transparency concerns, xAI published the full source code for Grok Build on GitHub under the Apache 2.0 license. The released codebase, consisting of approximately 844,530 lines of Rust, includes the agent loop, tools, terminal UI, and an extension system. While remnants of the upload function remain within the code, they are currently disabled. The tool has since been updated to support running entirely locally.
Evidence in the reporting
- Incident evidence
- SSH keys, password databases, documents, and photos were transferred
- Agent involvement
- uploaded all files in a directory to xAI's Google Cloud