What is Caution?¶

Understand what Caution is, why it exists, and how it changes the trust model for sensitive workloads.

What Caution is¶

Caution is a general-purpose, verifiable confidential compute platform for deploying sensitive applications in secure enclaves. It connects running enclave measurements back to the intended source code and build inputs that produced the enclave image. It currently supports deployments on AWS Nitro Enclaves, with additional attestation backends in active development.

It combines enclave isolation, hardware attestation, reproducible builds, and support for end-to-end encryption. Together, these let users verify code integrity and, when end-to-end encryption is enabled, keep application data encrypted all the way into the enclave.

Why it exists¶

Caution exists to move sensitive cloud services from "trust us" to "verify it yourself." It helps operators reduce blind trust by giving customers, auditors, and downstream services independent evidence of what is running. That matters when sensitive data, secrets, or automated decisions depend on software running outside the verifier's control.

The problem it solves¶

Cloud applications often require users to trust infrastructure operators, deployment pipelines, cloud providers, and private build systems. Confidential compute improves this by isolating workloads from the host and using hardware attestation to report enclave measurements.

That is an important baseline, but it does not answer what code is running inside the enclave. An operator or compromised deployment pipeline could run code that differs from the intended source code inside a genuine enclave, and isolation alone would provide no way to detect it.

Caution makes confidential compute verifiable by connecting enclave measurements back to the intended source code and build inputs that produced the enclave image.

It gives organizations, customers, auditors, and downstream services integrity guarantees that connect the intended source code, build inputs, and configuration to what is actually running inside an enclave. It also simplifies enclave deployments by integrating deployment and verification into familiar Git-based and CLI workflows.

Who it is for¶

Caution is for teams building services where ordinary cloud trust is not enough. As a general-purpose platform, Caution is not tied to one industry or application type; any workload that benefits from independently verifiable code integrity and enclave-protected data can use the model.

Example workloads that benefit from source-to-enclave verification include AI inference, oracles and data feeds, key management and signing systems, and nodes or staking infrastructure.

Caution is a fit when:

• Your customers or downstream services need to verify the software they interact with
• You want customers to depend less on infrastructure administrators
• You need to reduce insider risk from operators or compromised deployment pipelines
• You need stronger guarantees for code integrity, data confidentiality, or secret delivery
• You need deployment options that match your infrastructure, data residency, or control requirements

What makes it different¶

Most confidential compute systems can prove that a workload is running in a protected environment and that the deployed image has not changed since launch.

Caution is designed to turn confidential compute into verifiable compute by adding a source-to-enclave chain of evidence:

This lets verifiers evaluate the source and build process behind a running enclave, not only the binary measurements reported by hardware attestation.

Caution is fully open source, so teams can inspect the platform, understand the verification path, and self-host it if they want to operate deployments independently.

Caution turns verification into a developer workflow. Teams deploy with familiar Git-based workflows, expose attestation endpoints, and let verifiers reproduce enclave images locally.

Security model¶

Caution's security model is designed to reduce trust in operators, infrastructure, and deployment pipelines by combining four capabilities:

• Isolation: Workloads run inside confidential compute enclaves isolated from the host environment.
• Verifiability: caution verify checks that the running enclave matches the expected source and build inputs.
• Reproducibility: Deterministic, source-bootstrapped tooling lets enclave images be rebuilt and compared from the application down to the kernel.
• End-to-end encryption: When enabled, application data is encrypted all the way into the enclave, so infrastructure operators do not see plaintext.

These capabilities are strongest when the application source is available, the build is reproducible, the app runs outside debug mode, and end-to-end encryption is enabled for workloads that need to keep plaintext from the host.

This security model is grounded in the Distrust Threat Model, an adversary framework designed by the same team behind Distrust and Caution. The threat model assumes systems may already be compromised at some level. Caution applies that assumption by reducing reliance on operators, infrastructure, and deployment pipelines.

What Caution does not prove¶

Caution does not determine whether source code is safe, bug-free, or appropriate for a given use case. Verification proves that the running enclave matches the intended source code and build inputs. Users and auditors still need to inspect the code and decide whether they trust what it does.

Next steps¶