Caution

Teams handling sensitive workloads such as custody and signing, LLM inference over private data, staking infrastructure, oracles, and PHI need to prove their production systems are running exactly the code that was reviewed, not a tampered or forked version. Regular cloud environments require trusting the operator, and even confidential computing with TEEs only moves that trust to the hardware vendor while the software stack (application, runtime, OS, compiler) stays opaque. Without a way to verify what is actually running, regulators, customers, delegators, and auditors have to take claims about integrity on faith. Building this kind of verification infrastructure in-house would require significant platform engineering effort.

Caution is a verifiable compute platform built on top of AWS Nitro Enclaves that cryptographically links code running inside an enclave back to the reviewed source code, build inputs, and configuration that produced it. Developers use a Git-based workflow: `caution init` sets up the app, `git push caution main` builds a reproducible enclave image and provisions the TEE, and `caution verify` rebuilds the image, compares hashes, and confirms what is actually running via attestation. This extends hardware-level isolation from Nitro with software-level verification down to the compiler and toolchain layer. Caution offers bring-your-own-compute and fully managed deployment options, and since it is open source, teams can self-host it as well. The company states it plans to add support for Intel TDX, AMD SEV-SNP, and TPM 2.0 attestations in 2026.