xAI Open-Sources Grok Build After Data Exfiltration Scandal

Official Sources#

Last updated: July 15, 2026

Less than 48 hours after a security researcher documented Grok Build CLI uploading entire repositories - including .env files with secrets - to xAI's Google Cloud infrastructure, the company has released the full source code on GitHub. The timing is not lost on the developer community.

What's in the Release#

The grok-build repository contains the Rust source for xAI's terminal-based AI coding agent. According to the README:

• Full-screen TUI for interactive coding
• Agent runtime with shell command execution, file editing, and web search
• Support for headless/CI mode and the Agent Client Protocol (ACP)
• Builds for macOS, Linux, and Windows

The codebase is Apache 2.0 licensed with third-party notices acknowledging code ports from OpenAI's Codex and SST's opencode.

From the README:

This repository contains the Rust source for the grok CLI/TUI and its agent runtime. It is synced periodically from the SpaceXAI monorepo.

You can build it with cargo run -p xai-grok-pager-bin or install the release binary via the official installer.

The Context: Last Week's Data Scandal#

On July 13, a security researcher named cereblab ran Grok Build through mitmproxy and captured what it actually sends home. The findings were damning:

• 5.1 GiB uploaded when ~192 KB would have sufficed for model context
• Full git repositories including commit history uploaded to grok-code-session-traces GCS bucket
• .env files with credentials transmitted unredacted
• The "privacy toggle" in settings did nothing to prevent uploads

When a test .env file containing simulated credentials was placed in the working directory, it appeared verbatim in both live model requests and archived session uploads. The upload behavior persisted even when users explicitly instructed the CLI not to access certain files.

For full technical details, see our wire-level analysis coverage.

What HN is Saying#

The GitHub release drew 90+ points and 100+ comments on Hacker News. The discussion is deeply skeptical.

The cynical read:

The most common take is that this is pure damage control. One top comment: "I wonder if releasing this may have been on the roadmap, but been prioritized as a bit of whiplash following the 'you forfeit the entirety of your working directory as a condition of working with this tool' upset."

Another commenter was more direct: "xai is now in pure damage control mode, after they caught exfiltrating data from users."

Several commenters noted this feels like a tactical move rather than a principled open-source commitment: "If you have an LLM with less than 1% of the share to begin with, you suffer from bad rep and you got caught uploading user data, one of the very few remaining tactical moves to try to climb out of it is this."

The trust problem:

Even developers who praised the TUI quality are cautious. One wrote: "Really good TUI harness, it's a shame with the other news, but some other TUI agents should take some inspiration from pieces of this."

Others are staying away entirely: "Grok has had far too many instances where it's clear that the team building it cannot be trusted and does not care to build trustworthy products. I highly caution anyone from using any tools from xAi."

The technical interest:

Some commenters see value in the source for reverse-engineering purposes: "Trying to reverse engineer some specifics of how it does stuff has been a pain in the ass, and this will make it easier."

One developer asked whether the repo even compiles without the rest of the monorepo: "The commit message says 'initial sync from the monorepo.' Is this even compilable without the rest of the source code?"

The trace.rs file:

Multiple commenters zeroed in on the upload code at crates/codegen/xai-grok-shell/src/upload/trace.rs, asking whether this is the "infamous cloud upload routine." One noted: "I'm not sure it is indeed insidious, though it is of course possible that the code has been filtered out."

What xAI Changed#

According to reports, xAI has:

1. Disabled the upload feature server-side
2. Added a disable_codebase_upload config option in the CLI
3. Promised to delete previously uploaded data (per Elon Musk on X)

Whether any of this is verifiable from the open source is unclear. The trust model for a coding agent requires you to believe the binary you're running matches the source, and that the source actually represents all behavior.

The Ironic Model Quality Take#

Several commenters noted that Grok 4.5 is actually a good model. One wrote: "It's a shame that they exfiled private data. The model is actually good (better than opus 4.8 imo) and the harness itself is butter smooth with the potential of being the best out there."

This creates an awkward situation where the product quality is high but the trust is destroyed. As one commenter put it: "if you like grok-4.5 model, I suggest use the model directly via API, or use Grok's OAuth tokens if you are using supergrok+ subscriptions and connect it to your own agent."

Developer Guidance#

If you're considering Grok Build after the open-source release:

The case for: The TUI is genuinely well-designed. The Rust codebase is now auditable. If you build from source and run through a proxy, you can verify what it sends.

The case against: Trust is hard to rebuild. The company shipped code that uploaded user data without meaningful consent, and the privacy toggle was non-functional. The open-source release happened under pressure, not as a principled choice.

The middle ground: Use the model (Grok 4.5) directly through API if you need it, but route it through your own agent harness. Several open-source harnesses like Claude Code, Codex CLI, or opencode can connect to third-party models.

What This Means for Coding Agents#

Every coding agent has access to your filesystem. The implicit contract is that they process files locally or transmit only what's needed for model inference. Grok Build broke that contract by uploading entire repositories including files never touched during the session.

The open-source release sets a precedent that coding agents can be audited. But it also raises the bar: if your agent isn't open source, users have to trust your claims about data handling. After Grok Build, that trust is harder to earn.

For teams evaluating coding agents, this is a reminder to:

• Run agents through a proxy during evaluation
• Check what actually gets transmitted, not just what docs claim
• Prefer open-source agents where you can audit the code
• Assume anything the agent can access might be transmitted

The open-source release of Grok Build is a step toward accountability. Whether it's enough to rebuild trust depends on whether xAI's behavior changes, not just their codebase visibility.

Sources#

• xAI Grok Build GitHub Repository
• Hacker News Discussion on Open Source Release
• Cereblab Wire-Level Analysis
• xAI Response Coverage - DataBreaches.net
• The Agent Report Coverage