Skip to content

Trust But Verify: Security and Privacy Considerations for Remote Build Execution (RBE) Platforms

For organizations using Bazel and RBE, your build graph is effectively a fingerprint of your product ecosystem and the engineering hierarchy behind it. Depending on the extent to which Bazel and monorepo usage is established within your organization, your Bazel build graph may be composed of interdependent (configured) targets, actions, and artifacts spanning every single dependency, service and test that exists within your codebase.

Other highly valuable sources of metadata related to your codebase that Bazel can produce are JSON trace profiles, effectively structured recordings of what happened during a build and how long it took.

Combined, and observed in time, those two data sources offer insights into release cadence, acquisition targets, engineering headcount, and products potentially being shipped before you are actually ready to ship them. In other words, they are some of the most sensitive data sources an engineering organization produces, and they are routinely treated as exhaust.

Metadata, although generally uninteresting at glance, can certainly provide an advantage to malicious third parties looking to negatively impact your organization. For example, in early 2026 an error in an NPM registry file made the source code for Claude Code available to the public, exposing internal routing logic, API integrations, and communication pathways used by the agent to interface with cloud services and external endpoints. In a more famous example, the fitness tracker Strava's public features and default settings were exploited using Open-Source Intelligence (OSINT) to reveal the locations of secret military assets and government officials.

The Importance of Opt-In Telemetry

Unlike some other vendors in the Bazel space, EngFlow build tooling opens no connection you did not configure. We don't surprise our customers with usage beacons, anonymized metrics, or "help us improve the product" payloads, hidden behind some opt-out toggle and on "by default". The only endpoints your builds reach are the cluster endpoint you set and the infrastructure you already own.

That's because your cluster runs in your infrastructure. Everything is secured behind your network policy and your Identity and Access Management (IAM) framework regardless of deployment mode: in your AWS, GCP, or Azure account, on-prem, with Kubernetes, etc. This means that source code, artifacts, and execution logs stay inside those imposed boundaries. Plus, EngFlow staff hold no standing access. You decide when support access is granted and revoked (or the scope changes).

The client is Bazel, and EngFlow implements the open Remote Execution API. You point --remote_executor at your cluster and you are done. There is no EngFlow module in MODULE.bazel whose behavior we can change in a point release without you noticing. The one binary we do make available to developers is our auth CLI tool, which is open source: github.com/EngFlow/auth.

Privacy and Compliance Considerations

EngFlow maintains SOC 2 Type 2 compliance, and we provide the corresponding report under NDA upon request. Don't take our word for it, trust but verify. You can independently confirm our claims by restricting egress of your cluster endpoint and identify the infrastructure involved. Run a cold build while capturing existing network traffic using tools like tcpdump, eBPF, DNS or VPC flow logs. Finally, compare the resulting destination list against your configuration looking for a perfect match.

EngFlow's Commitment to Transparent Telemetry

We refuse to add hidden telemetry to EngFlow build tooling. If we ever request product usage data, we will ask directly, specify the exact fields, and ensure participation remains strictly optional.

If you find EngFlow software opening a connection you did not configure, kindly reach out to us over at security@engflow.com; the team will be happy to assist you.