All posts

SpaceX open sources Grok Build after repo-uploading scandal

Huma ShaziaJuly 18, 2026 at 2:02 PM5 min read
SpaceX open sources Grok Build after repo-uploading scandal

Key Takeaways

Privacy Disaster Pushes xAI to Open-Source Grok Build

SpaceX open sources Grok Build after repo-uploading scandal
Source: www.theregister.com
  • SpaceX released Grok Build's 844,530 lines of Rust code on GitHub after researchers discovered the tool was uploading user repos to Google Cloud
  • The company claims all retained coding data has been deleted and data retention is now off by default for all users
  • The open-source release lacks git history, making independent verification of past behavior difficult

SpaceX released Grok Build's source code on Wednesday, four days after security researchers at Cereblab discovered the AI coding tool was quietly uploading users' entire repositories to Google Cloud storage. The company's response includes promises to delete all stored data and disable data retention by default, but the rushed open-source release raises its own questions.

Advertisements

What did Grok Build actually do with user code?

Cereblab's analysis, published Sunday, showed Grok Build packaging repositories as Git Bundles and transmitting them to SpaceX-controlled cloud infrastructure. The behavior was not disclosed to users. Non-enterprise customers had data retention enabled by default, meaning their code was stored on company servers without explicit consent.

Elon Musk responded within days, a speed unusual for a company that rarely acknowledges external criticism. SpaceX, the parent company of xAI (where both X and Grok reside), committed to three actions: delete all retained data, change the default to no retention, and open-source the Grok Build CLI.

The open-source release is incomplete

The GitHub repository appeared Wednesday. Simon Willison, creator of Datasette and Django co-creator, counted 844,530 lines of Rust code. The code that sent repos to the cloud is still present but appears modified to reverse the behavior.

Here's the problem: it's a single commit. No pull requests, no git history. Developers cannot trace what changed between the version that uploaded their code and the version SpaceX now calls privacy-respecting. The company asks for trust while withholding the evidence that would justify it.

SpaceX's privacy claims deserve scrutiny

In its statement, SpaceX said Grok Build has "always respected zero data retention (ZDR)" for enterprise customers. That framing is careful. Enterprise customers got ZDR by default. Everyone else got the opposite. The company now says non-enterprise users "always had the ability to disable data upload," but burying an opt-out in CLI settings is not the same as informed consent.

SpaceX claims it disabled default retention on July 12 and is deleting all previously retained coding data. The company also invited security researchers to probe the tool and report issues through its bug bounty program, which pays $100 to $20,000 depending on severity.

The company's statement ends with a bold claim: "With these steps, Grok Build goes beyond other major coding products to protect user privacy." That's marketing, not fact. GitHub Copilot, Amazon CodeWhisperer, and other AI coding assistants have faced similar scrutiny, but SpaceX is the one that just got caught uploading repos without clear disclosure.

Advertisements

What this means for teams evaluating AI coding tools

Enterprise IT teams already treat AI coding assistants as a data governance risk. This incident reinforces that caution. Before adopting any AI dev tool, security teams should answer: Where does code go? Who can access it? What's the retention policy? Is it contractually enforceable?

SpaceX's response was fast, but speed doesn't equal trustworthiness. The company released source code without history, promised data deletion without third-party verification, and claimed privacy leadership while defending a default that silently uploaded proprietary code.

ℹ️

Logicity's Take

SpaceX did the minimum viable damage control. Open-sourcing the code is the right move, but stripping the git history undermines the transparency it's supposed to demonstrate. For CIOs, the lesson is not that Grok Build is uniquely bad. It's that every AI coding tool should be treated as a potential data exfiltration vector until proven otherwise. GitHub Copilot Business ($19/user/month) and Amazon CodeWhisperer (free tier available) both offer enterprise data controls, but none of these vendors should get the benefit of the doubt. Require contractual ZDR guarantees, not just UI toggles.

The bigger pattern in AI transparency

This is not the first time an AI company has promised self-regulation after getting caught. Grok Build's data grab follows a familiar script: deploy aggressively, apologize when exposed, and promise reforms that are difficult to verify. The pattern repeats because it works. Users return, regulators move slowly, and the data has already been collected.

SpaceX says you can now run Grok Build "fully open-sourced and local-first with your own inference." That's a meaningful option for teams willing to self-host. But most users will continue running the default configuration, trusting that the company has actually changed its behavior. That trust is not yet earned.

Frequently Asked Questions

What data did Grok Build upload without user consent?

Grok Build packaged entire Git repositories as Git Bundles and uploaded them to Google Cloud storage controlled by SpaceX. This affected non-enterprise users who had data retention enabled by default.

Has SpaceX actually deleted the retained code?

SpaceX claims it is deleting all previously retained coding data, but no third-party audit has verified this. Users must take the company at its word.

Can I run Grok Build locally without sending data to SpaceX?

According to SpaceX, yes. The open-source release allows users to run Grok Build locally with their own inference, avoiding any data transmission to company servers.

Why is the Grok Build source code a single commit with no history?

SpaceX has not explained this. The lack of git history prevents independent verification of what code was changed between the version that uploaded repos and the current release.

Also Read
DeepMind CEO wants AI self-regulation. Analysts aren't sold

Related analysis on AI industry self-regulation promises

ℹ️

Need Help Implementing This?

If your team is evaluating AI coding assistants and needs help building a security review framework, contact Logicity's consulting partners for vendor-neutral guidance on data governance and AI tool procurement.

Source: www.theregister.com

H

Huma Shazia

Senior AI & Tech Writer

Produced with AI assistance and reviewed by the Logicity editorial team. Learn more in our Editorial Policy.