xAI has released a security FAQ for Grok Bot, detailing how the system addresses common security concerns. The document, titled "Grok Bot security FAQ," provides answers to questions frequently posed by security reviews, covering areas such as isolation, network and egress controls, approval mechanisms, logging, and data handling.
Key Points
- Each Grok Bot user receives a dedicated Firecracker microVM with its own kernel, memory, and virtual devices, isolating users from one another.
- Within a single user's environment, all Bots share the same computer, meaning any login or file on that computer is accessible to every Bot run by that user.
- A Bot has no independent identity and cannot hold more access than the signed-in member; connector tokens remain on Cursor's backend.
- Blocking a plugin does not block access to that service's website, as connector policy and network policy are distinct layers.
- Network Controls, which enable administrators to set a destination allowlist for Grok Bot computers, are an Enterprise only feature.
- Grok Bot uses shared static egress IP addresses, which are available from account teams, and dedicated per-customer IPs are not provided.
- Auto Review evaluates risky Bot actions such as shell commands, plugin calls, and automation writes, and can allow, require approval for, or deny an action.
- Audit logs, Action Recording, and OpenTelemetry Export are Enterprise only features for logging and monitoring Grok Bot activities.
- Grok Bot runs exclusively on Cursor-hosted cloud computers, with on-premises deployment or bring-your-own-image deployment not supported.
- Anysphere, the entity behind Grok Bot, holds ISO/IEC 27001 and ISO/IEC 42001 certifications, with Grok Bot included in the current ISO scope.
Context
According to xAI, the security FAQ complements the "Grok Bot security" documentation, which outlines controls and deployment details for network policy, egress, approvals, identity, logging, data handling, hosting, and certifications. The FAQ specifically addresses the questions security reviews ask most often, while the security page provides the underlying controls. Dashboard settings and rollout steps are detailed in separate documentation for "Grok Bot for teams and enterprises."
Why It Matters
Builders and organizations deploying Grok Bot must understand its security architecture and available controls to manage risks effectively. The distinction between features available to Enterprise customers versus self-serve Teams impacts how organizations can implement network policies, logging, and administrative oversight, directly affecting their security posture and compliance efforts.
What To Do
- Review the "Grok Bot security FAQ" and "Grok Bot security" documentation to understand the isolation mechanisms and access controls.
- Note which security features, such as Network Controls, audit logs, and Action Recording, are designated as Enterprise only.
- Compare the available logging and monitoring options, including OpenTelemetry Export, with your organization's existing security information and event management (SIEM) systems.
- Contact your account team for current static egress IP ranges and to discuss any contractual restrictions regarding subprocessors or written residency commitments.