Security

Your data never leaves
the vault.

Zero-trust architecture, isolated execution, and operator-grade auditability for teams that need AI employees to behave like real production systems.

SOC 2 Type IIISO 27001GDPRCCPA

Four layers of defense

Zero-trust gateway

mTLS · RBAC · Rate limiting

Every request is authenticated and authorized. No implicit trust path exists between customer systems and agent runtimes.

Isolated compute

Dedicated VMs · No shared memory

Each agent gets its own sandboxed environment, reducing blast radius and eliminating lateral movement between tenants.

HSM-backed encryption

AES-256 · TLS 1.3 · Key rotation

Credentials stay encrypted in transit and at rest, with managed key rotation and vault-backed storage controls.

Immutable audit trails

Reasoning traces · Exportable logs

Every decision, tool call, and data access is logged for operator review, investigations, and enterprise assurance workflows.

1Request
2Auth
3Sandbox
4Execute
5Log

Security FAQ

Can agents access other customers' data?

No. Each agent runs in full isolation with separate runtime boundaries, credentials, and storage controls.

Do you train on my data?

No. Customer data is used to execute configured workflows and is not used to train or fine-tune models.

What happens if an agent needs human oversight?

Operators can require approvals for high-stakes actions and review a full log of what the agent observed, decided, and attempted.

Can enterprise teams request a deeper security review?

Yes. Clawly supports trust reviews, architecture walkthroughs, and questionnaire follow-up for enterprise procurement.

Need a deeper security review?

We can support procurement reviews, architecture walkthroughs, and follow-up questions for your security team.