Zero-trust
AI inference,
cryptographically
guaranteed.
Enterprise prompts encrypted client-side, decrypted only inside an attested AWS Nitro Enclave. No plaintext ever reaches the model provider, infrastructure owner, or cloud operator.
Your prompts
are not
private.
Every regulated enterprise that wants to use modern LLMs hits the same wall: the model provider, the cloud operator, and your own infrastructure team can read every prompt. There is no cryptographic guarantee otherwise, until now.
Four stages.
Each one
provable.
A single request passes through four cryptographically chained stages. Each stage is verifiable independently. None requires trust.
SDK encrypts prompts client-side using the enclave's public key. Plaintext never leaves your browser unencrypted.
AWS Nitro produces a signed attestation document with PCR measurements proving the enclave is running unmodified code.
Inside the enclave: decrypt, redact PII, run inference, re-encrypt. No process outside the enclave sees plaintext.
Each request returns an audit receipt with an attestation hash. Anyone can verify it on the public verifier.
Inside the
enclave boundary.
A vsock-bridged pipeline. Every layer outside the boundary handles ciphertext only. The boundary itself is cryptographically attestable end-to-end.
Usage-based. No lock-in.
All tiers include cryptographic attestation, PII redaction, and the full audit pipeline. You pay for inference volume, not features.
- Shared enclave (us-east-1)
- Claude 3.5 Sonnet · Llama 3.1 70B
- Public attestation verifier
- Community support
N/08From the Field“We could not ship our GenAI pipeline to legal review until Latensgate. The attestation receipts moved the conversation from trust to proof.”
Yasmin KarimovVP Platform Security · Atlas Capital
