
Caution is a Cayman-based verifiable compute platform that proves what runs in production.
Caution is a verifiable confidential compute platform for teams that cannot afford to have their software hacked.
Customers deploy workloads into secure enclaves and receive cryptographic proof that production runs the exact reviewed source code, build, and configuration, verified down to the kernel rather than only at the application layer. The platform is open source, runs on AWS Nitro Enclaves today, and is offered as fully managed, bring-your-own-compute, or self-hosted deployments.
Caution positions verifiable compute as a compliance and trust requirement rather than an optional hardening step, pointing to frameworks such as the EU AI Act as the shift that turns proof of system behaviour into a precondition for selling into regulated markets.
The company targets AI, fintech, digital assets, and healthcare as its first markets and says it is expanding toward multi-hardware attestation, additional cloud backends, and GPU workloads for AI inference. Its stated ambition is to make verifiable compute as default for network services as TLS became for transport.
Caution pairs reproducible builds with multi-hardware attestation, so a deployed enclave image can be rebuilt independently from source and compared against live attestation rather than trusted.
It also abstracts the enclave workflow behind a single git-driven pipeline that runs across enclave platforms, which replaces weeks of bespoke infrastructure engineering and removes dependence on any one cloud or hardware root of trust. The entire platform is open source under AGPLv3, so customers can self-host or audit every component instead of trusting the vendor.
Caution currently supports AWS Nitro Enclaves only, with Intel TDX, AMD SEV-SNP, and TPM 2.0 attestation still listed as in development rather than generally available.
The company is early-stage and entering its first revenue period, and its broader multi-cloud and bare-metal support is described as roadmap rather than shipped capability. The service is also positioned for high-assurance buyers who need to prove integrity, which is a narrower entry point than incumbent cloud confidential-compute offerings that ship with the platform a customer already runs on.
Caution prices by deployment model rather than by published seat or compute tiers, directing buyers to talk to its engineering team for terms.
The catalogue lists bring-your-own-compute for teams that want Caution managing enclaves inside their own AWS account, fully managed hosting for the fastest path to production, and a self-hosted path under AGPLv3 or a paid commercial licence for organisations that want to remove those obligations. Fees are invoiced in United States dollars, with optional paid support packages sold separately.