Skip to content

FAQ for security reviewers

Do you hold any certification or authorization?

Section titled “Do you hold any certification or authorization?”

No. Not FedRAMP, GovRAMP, CMMC, SOC 2, ISO 27001, HIPAA, PCI DSS or any DoD impact level. Those are targets. The trust center shows our self-assessed control mapping, generated from OSCAL.

No. Trials are not open to the public; the platform runs in our lab.

In your cluster, in the provider and region you choose (residency.placements in the assurance profile), or on your own hardware. Each customer has a dedicated cluster; nothing is shared with another tenant. Profiles can require US persons only for our platform staff, and IL5/IL6 profiles are restricted to AWS GovCloud, Azure Government or your hardware. Our own staff access is break-glass by default, and can be none. Status: In development.

You choose (keyCustody.mode): our management OpenBao (platform-transit, only allowed for commercial NIST Low/Moderate profiles), your cloud KMS (bring or hold your own key), your HSM over PKCS#11 or KMIP, or Shamir shares held by your named custodians with no auto-unseal. The same root protects your secrets store, GitOps secret decryption and infrastructure state. Status: In development; our lab uses a static seal.

That is a design target: an air-gapped profile allows no tunnel and no egress, needs your own registry, HSM or Shamir custody, no cloud attestors and no public transparency log, and runs only on your hardware. Software is delivered as bundles (Zarf, Hauler). Task records verify offline today. Status: Planned; ingress for air-gapped sites is an open design item.

No. Nothing we run is FIPS-validated today; our Secure Boot keys and secrets store use non-validated modules. A profile can require FIPS 140-3 (compliance.fips140_3Required), which forces custody into your validated KMS or HSM and runs Keycloak in FIPS strict mode, but that alone does not make the whole system validated.

How are agents identified and constrained?

Section titled “How are agents identified and constrained?”

One SPIFFE ID per agent, issued only after node and workload attestation, one sandbox per agent, and per-task permissions. Tasks reserved to licensed professionals produce drafts that a licensed human must approve; high-risk physical tasks are never offered. See the security model.

Each task produces a signed task record you can verify yourself. Audit logs go to sinks you choose (for example object-locked storage or your SIEM), retained as your profile requires.

Is the TPM attestation in your lab hardware-backed?

Section titled “Is the TPM attestation in your lab hardware-backed?”

No. The lab uses a software TPM. Hardware TPM attestation with an approved-vendor list is a profile option for customer hardware and is not yet tested.