SpaceX has open sourced the Grok Build CLI days after researchers caught the AI coding tool uploading entire user repositories to cloud storage without clear consent.
Grok Build, the AI coding tool from the Musk technology empire, is now open source. The release landed on GitHub just days after security researchers discovered the command-line tool was quietly packaging up users’ entire code repositories and sending them to company-controlled cloud storage, triggering a wave of criticism that pushed Elon Musk into rare public damage control.
How the Controversy Started
The issue came to light on Sunday, when researchers at Cereblab analyzed network traffic generated by Grok Build. Their findings were striking: the tool was bundling complete repositories as Git Bundles and transmitting them to Google Cloud storage controlled by the company.
For developers, that discovery raised immediate red flags. Source code often contains proprietary business logic, credentials, and sensitive project information. The idea that a coding assistant was shipping full repositories off local machines, without users clearly understanding it, spread quickly and drew heavy backlash across the developer community.
The reaction was strong enough that Musk issued a public statement alongside SpaceX and its technical staff. He promised the company would delete all data Grok Build had ever stored and give users far more control over how their information is handled. SpaceX sits at the top of the corporate structure that includes xAI, the division housing both X, the platform formerly known as Twitter, and the Grok AI business.
The Open Source Release
As part of that response, Musk pledged that Grok Build would be open sourced once the codebase had been audited for security vulnerabilities, a move intended to rebuild trust in the product. On Wednesday, the company followed through. The code is now publicly available on GitHub under the xai-org organization.
There is a notable caveat. The repository was published as a single commit, meaning there is no git history and no pull requests. Anyone hoping to trace how the code evolved over time, or to see exactly when the repository-upload behavior was introduced or modified, will not find that record in the public release.
Announcing the release, SpaceX said it had open sourced Grok Build and reset usage limits for all users, adding that opening the code allows anyone to help make the tool a reliable and robust harness.
Simon Willison, the creator of Datasette and co-creator of Django, examined the release and reported that the codebase contains 844,530 lines of Rust. He also noted that the code responsible for sending repositories to the cloud is still present, but it appears to have been altered so that the behavior is reversed.
What SpaceX Says About Data Retention
In a separate statement published alongside the open source announcement, the company addressed the privacy questions directly. It said Grok Build has always respected Zero Data Retention, known as ZDR, which was applied by default for enterprise customers. However, it acknowledged that during the early beta, data retention was switched on by default for everyone else.
According to the company, users always had the option to disable data upload in the CLI, and that choice was honored when selected. The problem was the default setting, which meant most non-enterprise users had their data retained unless they actively opted out. Based on user feedback, that default has now been changed.
The company outlined several concrete steps it is taking:
- Retention off by default: Default data retention was disabled for all Grok Build users starting on July 12.
- Full data deletion: All coding data that was previously retained is being deleted.
- Open source harness: The CLI code is public, so anyone can inspect what the tool actually does.
- Local-first option: Users can run Grok Build fully open source and local-first with their own inference.
The company claims that with retained data deleted, retention switched off by default, and an open source harness, Grok Build now goes further than other major coding products in protecting user privacy.
Before and After the Backlash
| Area | Before | After |
|---|---|---|
| Data retention default | Enabled for non-enterprise users in early beta | Disabled for all users from July 12 |
| Previously stored data | Retained on company cloud storage | Being deleted for all users |
| Codebase visibility | Closed source | Open source on GitHub as a single commit |
Security Researchers Invited to Dig In
The company is also encouraging outside scrutiny. Researchers have been invited to probe Grok Build for security issues and report findings through its bug bounty program, which pays rewards ranging from $100 to $20,000 depending on the severity of the vulnerability.
That invitation matters because the open source release gives researchers far more to work with. With 844,530 lines of Rust now public, independent experts can verify whether the privacy promises match the code, something that was impossible when the tool was closed.
Why This Matters for Developers
The episode is a reminder that AI coding assistants sit in a uniquely sensitive position. They have direct access to source code, and the defaults they ship with determine what happens to that code. For most users, defaults are destiny, and few developers read every setting before running a new CLI tool.
The open source release is a meaningful step toward transparency, though the absence of git history means questions about the tool’s past behavior cannot be fully answered from the public code alone. For now, developers who want maximum control can run Grok Build locally with their own inference, and everyone else benefits from retention being off by default. The bigger lesson applies across the industry: trust in AI developer tools depends on verifiable behavior, not just public statements.
Frequently Asked Questions
Security researchers at Cereblab discovered that Grok Build was bundling users' entire code repositories as Git Bundles and uploading them to company-controlled Google Cloud storage, with data retention enabled by default for non-enterprise users during the early beta.
The company disabled default data retention for all users starting July 12, committed to deleting all previously retained coding data, reset usage limits, and open sourced the Grok Build CLI so anyone can inspect the code. Users can also run it locally with their own inference.
No. The GitHub release was published as a single commit, so there is no git history and no pull requests. The current code is visible, including the altered upload code noted by Simon Willison, but past changes cannot be traced from the public repository.
